{"data":[{"id":"abel","title":"Abel","slug":"abel","oneLine":"A typed AI agent framework that makes multi-step LLM pipelines debuggable, composable, and auditable.","thesis":"Naive prompt chaining is opaque by design: state lives in strings, control flow is implicit, and debugging requires replaying the entire chain. Abel replaces this with a typed execution graph where each agent node declares its input schema, output schema, prompt template, and fallback policy — making pipelines inspectable at every step.","status":"prototype","domains":["AI Agents","LLM Infrastructure","Developer Tools"],"roles":["Architect","Lead Engineer"],"stack":["Python","TypeScript","OpenAI API","Anthropic API","Pydantic","FastAPI","Redis"],"skills":["agent-systems","llm-orchestration","api-design","python","typescript"],"featured":true,"order":1,"metrics":[{"id":"abel-m-1","label":"Node Types Implemented","value":"12","unit":"node types","context":"Tool, retrieval, synthesis, decision, loop, fork, join, guard, map, filter, cache, validator","evidenceId":"abel-ev-2"},{"id":"abel-m-2","label":"LLM Backends Supported","value":"3","unit":"providers","context":"OpenAI, Anthropic, local Ollama — via unified adapter (local benchmark)"},{"id":"abel-m-3","label":"State Isolation","value":"100%","unit":"immutable context","context":"Each node receives a read-only context dict; state mutations are tracked as delta objects (architectural property)"}],"claims":[{"id":"abel-claim-1","statement":"Abel's typed DAG makes agent state inspectable at every step, eliminating the need to replay full pipelines to debug a single node failure.","evidenceIds":["abel-ev-1","abel-ev-2"],"audienceModes":["engineer","ai"]},{"id":"abel-claim-2","statement":"A unified adapter interface makes switching between OpenAI, Anthropic, and local Ollama a one-line config change — no node rewrites required.","evidenceIds":["abel-ev-3"],"audienceModes":["engineer","ai","recruiter"]},{"id":"abel-claim-3","statement":"Explicit fallback nodes with access to the full error payload enable graceful degradation without ad-hoc try/except logic scattered across the pipeline.","evidenceIds":["abel-ev-2"],"audienceModes":["engineer"]}],"evidence":[{"id":"abel-ev-1","kind":"diagram","label":"Agent Execution Graph","description":"Typed DAG showing how agent nodes declare inputs/outputs and pass structured state through a pipeline, with visible fallback paths.","verified":true,"displayStatus":null,"url":null},{"id":"abel-ev-2","kind":"repo","label":"Core Engine Source","description":"Python implementation of the graph executor: topological walker, Pydantic schema validation, retry/fallback logic, and Redis state backend.","verified":true,"displayStatus":null,"url":null},{"id":"abel-ev-3","kind":"writeup","label":"Adapter Interface Design Doc","description":"Design rationale for the unified LLM adapter protocol: `async def complete(messages, schema) -> schema`. Provider differences are fully encapsulated.","verified":true,"displayStatus":null,"url":null}],"artifacts":[{"id":"abel-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Abel","description":"Core framework source"},{"id":"abel-art-2","label":"Architecture Diagram","kind":"diagram","description":"Agent graph execution model"}],"audienceSummaries":[{"mode":"recruiter","headline":"Built a production-grade AI agent framework from scratch","bullets":["Designed a typed execution graph for multi-step LLM workflows","Supports OpenAI, Anthropic, and local models via a unified adapter pattern","Makes AI pipelines debuggable, auditable, and testable — not just functional"],"cta":"Read the architecture"},{"mode":"engineer","headline":"Typed DAG executor with Pydantic schema validation and pluggable LLM adapters","bullets":["Graph nodes declare typed input/output schemas validated at every boundary","Immutable context propagation with per-node retry/fallback policies","Redis-backed distributed state; topological execution with async fork/join"],"cta":"Read case study"},{"mode":"ai","headline":"Structured agent orchestration with provider-agnostic LLM routing","bullets":["Typed DAG replaces implicit prompt chains — control flow is explicit and inspectable","Multi-provider adapter with unified `complete(messages, schema) -> schema` interface","Conditional branching, parallel execution, and chain-of-thought fallback policies"],"cta":"View architecture"}]},{"id":"vidcrawl","title":"VidCrawl","slug":"vidcrawl","oneLine":"A video intelligence pipeline that turns YouTube channels into semantically queryable knowledge datasets.","thesis":"Video is the world's largest unstructured knowledge base, but it's completely unsearchable. VidCrawl converts YouTube channels into queryable datasets through a five-stage pipeline: transcript extraction (with Whisper fallback for uncaptioned videos), semantic chunking with boundary detection, OpenAI embedding, Pinecone vector indexing, and a FastAPI retrieval layer. The core insight: naive sentence splitting loses topic coherence in long-form content — semantic chunking with cosine-similarity boundary detection fixes this.","status":"prototype","domains":["Data Engineering","NLP","Knowledge Extraction","AI"],"roles":["Architect","Lead Engineer"],"stack":["Python","YouTube Data API","Whisper","OpenAI Embeddings","Pinecone","FastAPI","PostgreSQL"],"skills":["data-pipelines","embeddings","nlp","python","api-design","vector-search"],"featured":true,"order":2,"metrics":[{"id":"vc-m-1","label":"Processing Speed","value":"<3","unit":"min / hour of video","context":"Local benchmark — prototype infrastructure, not production-optimized","evidenceId":"vc-ev-1"},{"id":"vc-m-2","label":"Channels Crawled","value":"50+","unit":"YouTube channels","context":"Technical, finance, and science content — prototype scope"},{"id":"vc-m-3","label":"Chunks Indexed","value":"200k+","unit":"semantic chunks","context":"Prototype run — not a sustained production index"},{"id":"vc-m-4","label":"Whisper Coverage Savings","value":"~70%","unit":"compute reduction","context":"Qualitative estimate — Whisper runs only on uncaptioned videos (~30% of corpus)"}],"claims":[{"id":"vc-claim-1","statement":"The full pipeline — transcript fetch, semantic chunking, embedding, and vector upsert — processes a 60-minute video in under 3 minutes end-to-end.","evidenceIds":["vc-ev-1"],"audienceModes":["engineer","ai","recruiter"]},{"id":"vc-claim-2","statement":"Semantic chunking with cosine-similarity boundary detection outperforms fixed-size sentence splitting on retrieval accuracy for long-form video content.","evidenceIds":["vc-ev-2"],"audienceModes":["ai","engineer"]},{"id":"vc-claim-3","statement":"Hybrid transcription — YouTube captions first, Whisper fallback — achieves near-100% coverage while cutting Whisper compute cost by ~70%.","evidenceIds":["vc-ev-1"],"audienceModes":["engineer","ai"]}],"evidence":[{"id":"vc-ev-1","kind":"benchmark","label":"Pipeline Latency Benchmark","description":"End-to-end timing for transcript fetch, chunking, embedding (batched), and Pinecone upsert on a 60-minute video. Local benchmark on prototype infrastructure.","verified":true,"displayStatus":null,"url":null},{"id":"vc-ev-2","kind":"writeup","label":"Chunking Strategy Comparison","description":"A/B comparison of naive sentence splitting vs. semantic-overlap chunking on MRR@5. Prototype evaluation on 20 hand-annotated queries across 5 videos.","verified":true,"displayStatus":null,"url":null}],"artifacts":[{"id":"vc-art-1","label":"Pipeline Diagram","kind":"diagram","description":"Data flow from YouTube API to vector store"},{"id":"vc-art-2","label":"Query API","kind":"api","description":"FastAPI endpoint for semantic video search"}],"audienceSummaries":[{"mode":"recruiter","headline":"Built a scalable video intelligence pipeline from scratch","bullets":["Five-stage pipeline: transcript extraction → semantic chunking → embedding → indexing → retrieval","Processed 200k+ semantic chunks across 50+ YouTube channels (prototype)","Sub-3-minute end-to-end processing per hour of video content"],"cta":"See the pipeline"},{"mode":"ai","headline":"End-to-end multimodal knowledge extraction with semantic chunking","bullets":["Whisper transcription fallback + cosine-similarity boundary detection for topic-aware chunking","OpenAI batch embedding pipeline with configurable overlap and retry logic","Pinecone namespace partitioning with PostgreSQL metadata for filtered retrieval"],"cta":"Read architecture"},{"mode":"engineer","headline":"Async ETL pipeline with semantic chunking and vector storage","bullets":["Hybrid transcription: YouTube captions first, Whisper fallback (saves ~70% compute)","Sentence-transformer boundary detection over fixed-size splitting for retrieval quality","Token-bucket rate limiting, batch embedding, Pinecone namespace partitioning"],"cta":"Read case study"}]},{"id":"kalsher","title":"Kalsher","slug":"kalsher","oneLine":"A modular options backtesting system with separated signal, execution, and attribution engines.","thesis":"Most retail options backtesting tools conflate strategy logic with execution assumptions, producing Sharpe ratios that don't survive contact with real fill prices. Kalsher separates signal generation, position sizing, execution modeling (with realistic bid/ask spread), and performance attribution into four distinct, independently testable modules — making it possible to actually measure where edge comes from, and where it disappears.","status":"prototype","domains":["Quantitative Finance","Backtesting","Options","Data Engineering"],"roles":["Quant Developer","Lead Engineer"],"stack":["Python","Pandas","NumPy","Polygon.io","CBOE Data","Plotly","FastAPI"],"skills":["quant-finance","backtesting","options","python","data-pipelines","statistical-analysis"],"featured":true,"order":3,"metrics":[{"id":"kal-m-1","label":"Option Strategies","value":"8","unit":"implemented","context":"Credit spreads, iron condors, calendars, straddles, ratio spreads, 0DTE scalps, covered calls, protective puts","evidenceId":"kal-ev-1"},{"id":"kal-m-2","label":"Historical Data","value":"5+","unit":"years","context":"OHLCV + options chain via Polygon.io; CBOE volatility surface data"},{"id":"kal-m-3","label":"VIX Regimes Tracked","value":"4","unit":"regimes","context":"Low (<15), Medium (15-25), High (25-40), Crisis (>40) — every bar is tagged"},{"id":"kal-m-4","label":"Fill Model Realism","value":"Configurable","unit":"slippage model","context":"Bid/ask midpoint + fraction of spread; configurable per-strategy (local benchmark)"}],"claims":[{"id":"kal-claim-1","statement":"Kalsher's strict module boundary (SignalEngine → SizingEngine → ExecutionEngine → AttributionEngine) allows swapping signal logic without touching execution or risk code.","evidenceIds":["kal-ev-1"],"audienceModes":["engineer","quant"]},{"id":"kal-claim-2","statement":"Backtests on SPX 0DTE credit spreads show measurable edge degradation when execution timing is modeled with perfect fills vs. realistic bid/ask spread — a difference that naive tools ignore.","evidenceIds":["kal-ev-2"],"audienceModes":["quant"]},{"id":"kal-claim-3","statement":"Regime tagging on every bar makes it measurable that most tested strategies have positive expected value only in low-VIX regimes — a result that aggregated metrics obscure.","evidenceIds":["kal-ev-2"],"audienceModes":["quant","engineer"]}],"evidence":[{"id":"kal-ev-1","kind":"diagram","label":"Four-Module Architecture Diagram","description":"System diagram showing the clean boundary between SignalEngine, SizingEngine, ExecutionEngine, and AttributionEngine — each with a defined protocol interface.","verified":true,"displayStatus":null,"url":null},{"id":"kal-ev-2","kind":"benchmark","label":"SPX 0DTE Backtest Results","description":"Local benchmark: comparison of credit spread P&L with perfect fills vs. realistic spread modeling, segmented by VIX regime. Illustrates edge degradation and regime sensitivity.","verified":true,"displayStatus":null,"url":null}],"artifacts":[{"id":"kal-art-1","label":"Backtest Engine","kind":"repo","url":"https://github.com/DanielBoehmeIII/Kalsher","description":"Core Python backtesting engine"},{"id":"kal-art-2","label":"Performance Dashboard","kind":"demo","url":"https://kalsher.vercel.app/","description":"Plotly-based P&L and Greeks dashboard"}],"audienceSummaries":[{"mode":"quant","headline":"Modular options backtest engine with regime-aware attribution","bullets":["Four-module architecture: signal, sizing, execution, attribution — each with typed protocol interfaces","VIX regime tagging on every bar; attribution segmented by regime, strategy, and leg","8 option strategies including 0DTE credit spreads with realistic fill modeling"],"cta":"Read system design"},{"mode":"recruiter","headline":"Built a quantitative trading system with rigorous, auditable architecture","bullets":["Modular design separates signal logic from execution assumptions — enabling real A/B testing","Supports 8 option strategies, 5+ years of market data, regime-segmented performance analytics","Realistic fill modeling produces Sharpe ratios that survive contact with real spreads"],"cta":"See the system"},{"mode":"engineer","headline":"Protocol-based backtesting system with immutable bar state","bullets":["Typed protocol interfaces for each module — swap signal logic without touching execution code","Immutable PortfolioState between bars prevents all forms of lookahead bias","Post-hoc attribution engine decomposes P&L by regime, signal type, holding period, and leg"],"cta":"Read case study"}]},{"id":"theib","title":"Theib","slug":"theib","oneLine":"An AI-powered implementation scout that automates pre-coding research — finding, scoring, and synthesizing reference implementations.","thesis":"Engineers routinely skip implementation research because the manual process (search → read → evaluate → synthesize) takes 1-2 hours per feature. Theib automates this loop as a Claude Code slash command: given a feature description, it expands the query into search variants, fetches sources from GitHub, Semantic Scholar, and the web in parallel, scores each by license safety and quality, extracts reusable patterns from the top sources, and produces a structured dossier with an implementation plan and a ready-to-use next Claude Code prompt. Zero setup; runs in-context.","status":"shipped","domains":["Developer Tools","AI","Knowledge Systems","Research Automation"],"roles":["Architect","Lead Engineer"],"stack":["TypeScript","Claude API","GitHub Search API","Semantic Scholar API","Claude Code SDK"],"skills":["llm-orchestration","agent-systems","typescript","api-design","knowledge-systems"],"featured":true,"order":4,"metrics":[{"id":"theib-m-1","label":"Research Time Replaced","value":"60–120","unit":"minutes (est.)","context":"Qualitative estimate — based on typical manual research session timing for a feature with multiple implementation approaches","evidenceId":"theib-ev-1"},{"id":"theib-m-2","label":"Sources Scored Per Run","value":"20+","unit":"references","context":"Repos, papers, docs, and product references — scored in parallel before deep extraction"},{"id":"theib-m-3","label":"Pipeline Steps","value":"5","unit":"stages","context":"Query expansion → parallel search → scoring → pattern extraction → synthesis"}],"claims":[{"id":"theib-claim-1","statement":"Theib produces an actionable implementation dossier in under 2 minutes, replacing 1-2 hours of manual research — a qualitative estimate based on typical research session timing.","evidenceIds":["theib-ev-1"],"audienceModes":["recruiter","engineer","ai"]},{"id":"theib-claim-2","statement":"Automated license scoring filters GPL-contaminated and restrictively licensed sources before they reach the dossier, protecting downstream projects from inadvertent license contamination.","evidenceIds":["theib-ev-2"],"audienceModes":["engineer","recruiter"]},{"id":"theib-claim-3","statement":"Deploying as a Claude Code skill (not a standalone tool) eliminates setup friction — engineers use it in-context during development without switching workflows.","evidenceIds":["theib-ev-1"],"audienceModes":["engineer","recruiter","ai"]}],"evidence":[{"id":"theib-ev-1","kind":"demo","label":"Live Claude Code Skill Demo","description":"Theib running as the `/theib` slash command inside Claude Code, generating a real dossier for a feature request. Demonstrates end-to-end pipeline from query to structured output.","verified":true,"displayStatus":null,"url":null},{"id":"theib-ev-2","kind":"writeup","label":"License Safety Scoring Rubric","description":"Documented scoring rubric for license classification: permissive (MIT, Apache 2, BSD), weak-copyleft (LGPL, MPL), strong-copyleft (GPL, AGPL), and unknown. Sources with score < 0.5 are excluded from the dossier.","verified":true,"displayStatus":null,"url":null}],"artifacts":[{"id":"theib-art-1","label":"Claude Code Skill","kind":"repo","url":"https://github.com/DanielBoehmeIII/Theib","description":"Theib deployed as /theib slash command"},{"id":"theib-art-2","label":"Dossier Template","kind":"doc","description":"Structured output format for implementation dossiers"}],"audienceSummaries":[{"mode":"recruiter","headline":"Shipped an AI-powered developer tool actively used in engineering workflows","bullets":["Replaces 1-2 hours of manual research with a 2-minute slash command in Claude Code","Scores 20+ sources per run for license safety — protects downstream projects automatically","Zero setup: runs in-context as a Claude Code skill with no additional configuration"],"cta":"See the demo"},{"mode":"engineer","headline":"Multi-source research pipeline with structured scoring and LLM synthesis","bullets":["Parallel search across GitHub, Semantic Scholar, and web; 20+ sources per run","Rubric-based license and quality scoring — reproducible and auditable","Claude-synthesized dossier with a ready-to-use implementation plan and next prompt"],"cta":"Read case study"},{"mode":"ai","headline":"Agentic research pipeline deployed as an in-context Claude Code skill","bullets":["Five-stage pipeline: query expansion → parallel search → scoring → extraction → synthesis","Structured output at every stage via Claude API — consistent, parseable, testable","Broad-then-deep extraction strategy: score 20+ broadly, extract 3-5 deeply"],"cta":"Read architecture"}]},{"id":"nodiff","title":"NoDiff","slug":"nodiff","oneLine":"A research-stage semantic code review tool that understands intent, not just diffs, to surface non-obvious behavioral regressions.","thesis":"Standard diff-based code review shows what changed syntactically, not what broke behaviorally. A function can be refactored in a way that looks clean in the diff, passes all tests, and still violate the invariant it was supposed to maintain. NoDiff addresses this by combining Tree-sitter AST diffing (to identify structurally significant changes) with two-stage LLM reasoning: first extract the intent of the change, then check whether that intent is preserved. The key insight: asking 'what changed?' before 'what broke?' gives the model better grounding for regression analysis.","status":"research","domains":["Developer Tools","Code Intelligence","AI","Software Engineering"],"roles":["Researcher","Architect"],"stack":["TypeScript","Python","Tree-sitter","Claude API","GitHub API"],"skills":["code-intelligence","llm-orchestration","typescript","python","api-design"],"featured":true,"order":5,"metrics":[{"id":"nd-m-1","label":"Languages Supported","value":"4","unit":"languages","context":"TypeScript, Python, Go, Rust — via Tree-sitter grammars (implemented)"},{"id":"nd-m-2","label":"Per-PR Latency Target","value":"<30","unit":"seconds","context":"Prototype target — function-scoped analysis; full-file analysis would exceed practical PR workflow threshold"},{"id":"nd-m-3","label":"Case Study Regressions","value":"10","unit":"examples analyzed","context":"Planned evaluation: formal recall/precision measurement on larger corpus pending"}],"claims":[{"id":"nd-claim-1","statement":"Two-stage LLM reasoning (intent extraction before regression check) catches behavioral regressions that are invisible to test suites and line-level diff review — supported by a prototype case study of 10 real-world examples.","evidenceIds":["nd-ev-1"],"audienceModes":["engineer","ai"]},{"id":"nd-claim-2","statement":"AST-level diffing eliminates formatting noise from line-level diffs, producing a semantics-only ChangeSet that provides higher-signal context for LLM intent analysis.","evidenceIds":["nd-ev-2"],"audienceModes":["engineer"]},{"id":"nd-claim-3","statement":"Scoping analysis to changed functions (not full files) keeps per-PR latency under 30 seconds — a practical threshold for PR workflow integration.","evidenceIds":["nd-ev-2"],"audienceModes":["engineer","ai"]}],"evidence":[{"id":"nd-ev-1","kind":"writeup","label":"Regression Detection Case Study","description":"Prototype analysis of 10 real-world regressions: changes that passed all tests and looked clean in code review, but violated behavioral invariants detectable by two-stage LLM reasoning. Planned formal evaluation pending.","verified":false,"displayStatus":null,"url":null},{"id":"nd-ev-2","kind":"diagram","label":"AST Diff Pipeline Architecture","description":"Architecture diagram of the three-layer pipeline: structural layer (Tree-sitter AST diff), intent layer (Claude intent extraction), regression layer (Claude invariant check with chain-of-thought).","verified":true,"displayStatus":null,"url":null}],"artifacts":[{"id":"nd-art-1","label":"AST Diff Engine","kind":"repo","url":"https://github.com/DanielBoehmeIII/NoDiff","description":"Tree-sitter-based structural diff with intent annotation"}],"audienceSummaries":[{"mode":"engineer","headline":"AST-based semantic code review with two-stage LLM intent analysis","bullets":["Tree-sitter structural diff → intent extraction → regression check (chain-of-thought)","Catches behavioral regressions invisible to test suites and line-level diff review","Supports TypeScript, Python, Go, Rust; function-scoped for practical PR latency"],"cta":"Read the research"},{"mode":"ai","headline":"Two-stage LLM pipeline for semantic regression detection","bullets":["Intent extraction before regression check — grounding via explicit prior-state summary","AST-grounded context eliminates formatting noise from line diffs for higher-signal prompts","Chain-of-thought regression analysis with confidence scoring"],"cta":"See architecture"},{"mode":"recruiter","headline":"Building AI-powered code intelligence for semantic regression detection","bullets":["Research on a class of bugs invisible to tests: intent-layer regressions","Multi-language AST parsing with Tree-sitter; LLM intent analysis for code review","Prototype case study of 10 real-world regressions catches the two-stage approach works"],"cta":"Read more"}]},{"id":"agentic-system","title":"Self-Improving Agent","slug":"agentic-system","oneLine":"An autonomous evaluation loop that scores its own outputs, versions prompts, and rewrites workflows to measurably improve agent performance over time.","thesis":"Static LLM pipelines can't improve: prompts don't change, failure modes aren't systematically analyzed, and performance drift is invisible. This system implements a principled self-improvement loop: an evaluator agent scores each output against a structured rubric, failures trigger a prompt optimizer that generates candidates, and the best-performing version is committed with performance attribution. The key insight is that measurable improvement requires three things: a rubric, a versioned history, and a delta you can reason about.","status":"prototype","domains":["AI Agents","LLM Infrastructure","Evaluation Systems"],"roles":["Architect","Lead Engineer"],"stack":["Python","Claude API","Pydantic","SQLite"],"skills":["agent-systems","llm-orchestration","python"],"featured":false,"order":6,"metrics":[{"id":"ag-m-1","label":"Rubric Dimensions","value":"6","unit":"criteria","context":"Correctness, completeness, efficiency, style, edge-case handling, documentation — scored 0–10 per dimension"},{"id":"ag-m-2","label":"Prompt Versioning","value":"Full","unit":"with rollback","context":"Each version stored with performance delta and diff from parent — enables rollback to any checkpoint"}],"claims":[{"id":"ag-claim-1","statement":"The evaluator-optimizer loop measurably improves task performance over successive iterations — with versioned prompts and performance attribution showing exactly which change produced which gain.","evidenceIds":["ag-ev-1"],"audienceModes":["engineer","ai"]},{"id":"ag-claim-2","statement":"Structured rubric-based evaluation produces reproducible scores, making improvement traceable — not a subjective 'this feels better' assessment.","evidenceIds":["ag-ev-2"],"audienceModes":["engineer","ai","recruiter"]}],"evidence":[{"id":"ag-ev-1","kind":"writeup","label":"Evaluation Loop Architecture","description":"Design document for the evaluator-optimizer loop: rubric design, prompt versioning schema, performance attribution methodology.","verified":true,"displayStatus":null,"url":null},{"id":"ag-ev-2","kind":"benchmark","label":"Improvement Curves","description":"Performance curves across iterations on a code generation task. Local benchmark — shows measurable gain from baseline to optimized prompt versions.","verified":true,"displayStatus":null,"url":null}],"artifacts":[],"audienceSummaries":[{"mode":"ai","headline":"Autonomous evaluation loop with rubric-based scoring and prompt versioning","bullets":["Three-agent architecture: evaluator, optimizer, validator — each using structured output","6-dimension rubric produces reproducible scores; improvements attributed per-dimension","Versioned prompt store with performance deltas — every improvement traceable and reversible"],"cta":"Read architecture"},{"mode":"engineer","headline":"Self-improvement pipeline with testable evaluation and version-controlled prompts","bullets":["Rubric-based evaluation over LLM preference — reproducible, auditable, anti-gaming","3-candidate generation + held-out validation reduces overfitting to training task distribution","SQLite version store with attribution: which dimension, which change, how much improvement"],"cta":"Read case study"},{"mode":"recruiter","headline":"Built an autonomous agent that improves its own performance through structured evaluation","bullets":["Evaluator → optimizer → validator loop — fully autonomous with human-reviewable commit log","Structured rubric scoring (6 dimensions) enables traceable, attributable improvement","Addresses the real gap: most LLM pipelines can't systematically improve without evaluation infrastructure"],"cta":"Read more"}]},{"id":"cache-system","title":"Distributed Cache","slug":"cache-system","oneLine":"A Redis-inspired distributed key-value store built in C++ with LRU eviction, TTL expiration, persistence, and thread-safe concurrent access via shard-locked data structures.","thesis":"Most engineers use Redis without understanding what makes it hard to build correctly: concurrent access requires careful lock discipline, TTL expiration needs a background worker that doesn't block reads, and LRU eviction requires an O(1) structure that survives concurrent writes. This project implements all of this from scratch — TCP server, RESP-compatible protocol, 256-shard mutex concurrency, O(1) LRU via doubly-linked list + hashmap, write-ahead log persistence, and a background TTL expiry worker. The point isn't to replace Redis; it's to understand exactly how it works.","status":"prototype","domains":["Systems Programming","Distributed Systems","Infrastructure"],"roles":["Systems Engineer"],"stack":["C++","POSIX Threads","TCP/IP","Redis Protocol (RESP)"],"skills":["cpp","distributed-systems","systems-programming"],"featured":false,"order":7,"metrics":[{"id":"cs-m-1","label":"Mutex Shards","value":"256","unit":"shards","context":"Per-key assignment: hash(key) % 256 — ~0.4% of keys per shard, near-linear contention scaling"},{"id":"cs-m-2","label":"LRU Complexity","value":"O(1)","unit":"amortized","context":"Doubly-linked list + hashmap — O(1) get, put, and eviction without heap or array operations"},{"id":"cs-m-3","label":"RESP Commands","value":"12","unit":"implemented","context":"GET, SET, DEL, EXPIRE, TTL, EXISTS, KEYS, FLUSHALL, INCR, DECR, LPUSH, LPOP"}],"claims":[{"id":"cs-claim-1","statement":"Thread-safe concurrent access is achieved without a global lock: 256-shard mutex sharding provides O(1) access with contention bounded to ~0.4% of keys per shard under uniform key distribution.","evidenceIds":["cs-ev-1"],"audienceModes":["engineer"]},{"id":"cs-claim-2","statement":"LRU eviction runs in O(1) amortized time using a doubly-linked list + hashmap — maintaining eviction order without array shifts or heap operations.","evidenceIds":["cs-ev-1"],"audienceModes":["engineer"]}],"evidence":[{"id":"cs-ev-1","kind":"repo","label":"C++ Implementation","description":"Full C++ source: TCP server, RESP protocol parser, shard-locked concurrent hashmap, O(1) LRU doubly-linked list, TTL background worker, write-ahead log.","verified":true,"displayStatus":"private","url":null},{"id":"cs-ev-2","kind":"benchmark","label":"Throughput Benchmark","description":"Local benchmark: reads/sec and writes/sec under concurrent load (8 threads, 100k operations). Compared against naive single-threaded implementation.","verified":true,"displayStatus":null,"url":null}],"artifacts":[{"id":"cs-art-1","label":"C++ Source","kind":"repo","url":"https://github.com/DanielBoehmeIII/Distributed-Cache-System","description":"Thread-safe cache implementation (private)"}],"audienceSummaries":[{"mode":"engineer","headline":"Redis-inspired key-value store: shard-locked concurrency, O(1) LRU, WAL persistence — in C++","bullets":["256-shard mutex design: ~0.4% of keys per shard, near-linear throughput scaling","Doubly-linked list + hashmap LRU: O(1) get, put, eviction — no heap, no array shifts","Lazy TTL expiry on GET + 100ms background sweep — non-blocking on the request path"],"cta":"Read system design"},{"mode":"recruiter","headline":"Built a Redis-equivalent distributed cache from scratch in C++","bullets":["TCP server, RESP protocol, LRU eviction, TTL expiration, WAL persistence — all from scratch","Thread-safe concurrent access with 256-shard mutex design — no global lock","12 RESP-compatible commands, epoll event loop, configurable thread pool"],"cta":"Read case study"},{"mode":"ai","headline":"Low-level systems work: caching infrastructure in C++","bullets":["Foundational infrastructure knowledge: concurrency, networking, persistence, eviction","Complements AI system work — understanding the layer that agent state systems depend on"],"cta":"Read more"}]},{"id":"clipbounce","title":"ClipBounce","slug":"clipbounce","oneLine":"A video clip extraction and highlight generation pipeline for surfacing high-signal segments from long-form content.","thesis":"Long-form video contains high-value moments spread across low-density surrounding content. ClipBounce automates extraction of those moments using scene detection, transcript alignment, and configurable engagement scoring — producing highlight clips without manual scrubbing.","status":"prototype","domains":["Video Processing","Media Engineering","Content Tools"],"roles":["Architect","Lead Engineer"],"stack":["Python","FFmpeg","OpenCV","FastAPI"],"skills":["data-pipelines","python","api-design"],"featured":false,"order":8,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"cb-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/ClipBounce","description":"ClipBounce source"}],"audienceSummaries":[{"mode":"recruiter","headline":"Built a video highlight extraction pipeline with transcript-aligned scene detection","bullets":["Automates identification and extraction of high-signal clips from long-form video","Combines OpenCV scene detection with transcript alignment for topic-aware extraction","FFmpeg-based output with configurable scoring and duration constraints"],"cta":"View repo"},{"mode":"engineer","headline":"Transcript-aligned scene detection pipeline with configurable clip scoring","bullets":["OpenCV content-aware scene detection with configurable sensitivity","Two-pass approach: detect scenes, score independently for better selection quality","Precise FFmpeg timestamp cuts with audio-video sync preservation"],"cta":"View repo"}]},{"id":"kodek","title":"Kodek","slug":"kodek","oneLine":"A lightweight data encoding and format transformation utility with a unified interface for common serialization schemes.","thesis":"Working across system boundaries means constantly converting between JSON, YAML, TOML, CSV, and binary formats. Kodek provides a single CLI and library interface for encoding, decoding, and transforming data across these formats with optional schema validation — eliminating per-format tooling sprawl.","status":"prototype","domains":["Developer Tools","Data Engineering","Utilities"],"roles":["Lead Engineer"],"stack":["TypeScript","Node.js"],"skills":["typescript","api-design"],"featured":false,"order":9,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"kodek-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Kodek","description":"Kodek source"}],"audienceSummaries":[{"mode":"recruiter","headline":"Built a unified data encoding utility for cross-format transformation","bullets":["Single interface for JSON, YAML, TOML, and CSV encoding/decoding","CLI and library interfaces for both scripting and programmatic use","Optional schema validation on input and output"],"cta":"View repo"},{"mode":"engineer","headline":"Adapter-pattern format transformation library with composable CLI","bullets":["Format adapters implement a common encode/decode interface — no cross-format coupling","CLI composable with pipes; library usable without CLI dependencies","Round-trip fidelity tests per format on an edge-case corpus"],"cta":"View repo"}]},{"id":"multimodal-research-assistant","title":"Multimodal Research Assistant","slug":"multimodal-research-assistant","oneLine":"An AI research assistant that ingests PDFs, images, and text to answer research questions with cited, grounded responses.","thesis":"Researchers work with heterogeneous inputs — PDFs, diagrams, data tables, images, and text notes. A useful research assistant must handle all of these without requiring format conversion. This project builds a retrieval-augmented QA system that ingests mixed-modality inputs, extracts and indexes their content, and responds to research queries with citations back to the source material.","status":"prototype","domains":["AI","Knowledge Systems","Research Automation","NLP"],"roles":["Architect","Lead Engineer"],"stack":["Python","Claude API","OpenAI Embeddings","Pinecone","FastAPI","PyMuPDF"],"skills":["agent-systems","llm-orchestration","embeddings","python","knowledge-systems"],"featured":false,"order":10,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"mra-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Multimodal-Research-Assistant","description":"Multimodal Research Assistant source"}],"audienceSummaries":[{"mode":"recruiter","headline":"Built a multimodal AI research assistant with PDF, image, and text ingestion","bullets":["Retrieval-augmented QA system handling PDFs, images, and text without format conversion","Claude vision integration for image understanding and semantic indexing","Citation-grounded responses tied back to specific source documents and pages"],"cta":"View repo"},{"mode":"ai","headline":"Multimodal RAG pipeline with vision-augmented PDF ingestion and cited retrieval","bullets":["PyMuPDF text extraction + Claude vision for embedded image description","Pinecone vector store with modality and source metadata for filtered retrieval","Structured citation format enforced via Claude structured output"],"cta":"View repo"}]},{"id":"cloud-storage-service","title":"Cloud Storage Service","slug":"cloud-storage-service","oneLine":"An S3-inspired distributed object storage service with bucket management, multipart upload, and presigned URL access control.","thesis":"Building a cloud storage service from scratch makes explicit the design decisions behind distributed object storage: atomic write guarantees, efficient large file handling, access control that scales without per-request database lookups, and metadata indexing that handles prefix queries. This project implements core S3-equivalent functionality to internalize these problems.","status":"prototype","domains":["Distributed Systems","Infrastructure","Cloud Storage","Systems Programming"],"roles":["Systems Engineer"],"stack":["Python","FastAPI","PostgreSQL","Redis","Docker"],"skills":["distributed-systems","python","api-design","systems-programming"],"featured":false,"order":11,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"css-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Cloud-Storage-Service","description":"Cloud Storage Service source"}],"audienceSummaries":[{"mode":"engineer","headline":"S3-equivalent object storage with presigned URLs, multipart upload, and pluggable backends","bullets":["HMAC-signed presigned URLs for constant-time access control without DB round-trips","Atomic multipart upload assembly via filesystem rename","PostgreSQL metadata with GIN index for efficient prefix-based object listing"],"cta":"View repo"},{"mode":"recruiter","headline":"Built a cloud object storage service from scratch in Python","bullets":["S3-compatible API: bucket management, object CRUD, multipart upload","Presigned URL access control with HMAC signatures — no per-request database lookups","Pluggable storage backend: local filesystem or configurable cloud backend"],"cta":"View repo"}]},{"id":"job-queue-scheduler","title":"Scalable Job Queue Scheduler","slug":"job-queue-scheduler","oneLine":"A priority-aware distributed task queue with at-least-once delivery, retry policies, and worker health monitoring.","thesis":"Building a job queue from scratch exposes the design challenges that Celery and BullMQ solve but don't explain: at-least-once vs. exactly-once delivery tradeoffs, visibility timeout management, priority queue data structures under load, and worker heartbeat protocols for detecting silent failures.","status":"prototype","domains":["Distributed Systems","Infrastructure","Backend Engineering"],"roles":["Systems Engineer"],"stack":["Python","Redis","FastAPI","PostgreSQL","Docker"],"skills":["distributed-systems","python","api-design"],"featured":false,"order":12,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"jqs-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Scalable-Job-Queue-Scheduler","description":"Job Queue Scheduler source"}],"audienceSummaries":[{"mode":"engineer","headline":"Priority-aware job queue with visibility timeouts and worker heartbeat monitoring","bullets":["Redis sorted set for O(log n) priority-ordered enqueue/dequeue with timestamp tiebreaker","Visibility timeout pattern for at-least-once delivery without explicit acknowledgment","Heartbeat-based worker health: automatic re-queue on silent worker failure"],"cta":"View repo"},{"mode":"recruiter","headline":"Built a distributed task queue with at-least-once delivery and priority scheduling","bullets":["Redis-backed priority queue with visibility timeout for safe at-least-once execution","Worker health monitoring with automatic re-queue on worker failure","FastAPI interface for job enqueueing, status polling, and cancellation"],"cta":"View repo"}]},{"id":"options-vol-analyzer","title":"Options Volatility Surface Analyzer","slug":"options-vol-analyzer","oneLine":"A real-time options implied volatility surface analyzer with SVI parametric fitting, skew visualization, and vol regime detection.","thesis":"Options implied volatility surfaces encode the market's probabilistic view of future price distributions, but most retail tools display raw IV numbers in flat tables with no analytical layer. This project builds a live vol surface from options chain data, fits a parametric surface model, detects regime changes (flat vs. steep skew, contango vs. backwardation), and surfaces actionable signals for options positioning.","status":"prototype","domains":["Quantitative Finance","Options","Data Visualization","Real-Time Systems"],"roles":["Quant Developer","Lead Engineer"],"stack":["Python","NumPy","SciPy","Plotly","Polygon.io","FastAPI"],"skills":["quant-finance","options","python","statistical-analysis","data-pipelines"],"featured":false,"order":13,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"ovsa-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Real-Time-Options-Volatility-Surface-Analyzer","description":"Options Vol Surface Analyzer source"}],"audienceSummaries":[{"mode":"quant","headline":"Real-time options vol surface with SVI parametric fit and regime detection","bullets":["Black-Scholes IV inversion per strike/expiry with SVI surface parametrization","RBF interpolation for illiquid strikes; regime detection on skew slope and term structure","WebSocket live updates via Polygon.io options chain feed with Plotly 3D surface rendering"],"cta":"View repo"},{"mode":"recruiter","headline":"Built a real-time options volatility surface analyzer with live data feeds","bullets":["Computes and visualizes implied volatility across all strikes and expirations","SVI parametric model for smooth, arbitrage-free surface fitting","Live updates via WebSocket from Polygon.io options chain data"],"cta":"View repo"}]},{"id":"market-microstructure-alpha","title":"Market Microstructure Alpha Detector","slug":"market-microstructure-alpha","oneLine":"A real-time pipeline that extracts order flow imbalance, VPIN, and effective spread as short-horizon alpha signals from tick data.","thesis":"Market microstructure alpha — signals derived from order flow, bid/ask dynamics, and trade initiation patterns — decays in minutes to hours. Most quantitative tools operate at daily resolution and miss these signals entirely. This project builds a real-time pipeline that ingests tick-level trade and quote data, computes microstructure features, and generates short-horizon signals for mean-reversion and momentum strategies.","status":"prototype","domains":["Quantitative Finance","Market Microstructure","Algorithmic Trading","Real-Time Systems"],"roles":["Quant Developer","Lead Engineer"],"stack":["Python","NumPy","Pandas","Polygon.io","FastAPI","Redis"],"skills":["quant-finance","python","statistical-analysis","data-pipelines"],"featured":false,"order":14,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"mma-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Market-Microstructure-Alpha-Detector","description":"Market Microstructure Alpha Detector source"}],"audienceSummaries":[{"mode":"quant","headline":"Real-time microstructure signal pipeline: OFI, VPIN, effective spread, price impact","bullets":["Volume-time bucketing for stationary feature computation across market regimes","Lee-Ready trade initiation classification at sub-100ms latency","Redis signal store with TTL-based stale signal expiry; FastAPI consumption interface"],"cta":"View repo"},{"mode":"recruiter","headline":"Built a real-time market microstructure alpha detection system","bullets":["Computes order flow imbalance, VPIN, and effective spread from live tick data","Volume-time bucketing for regime-stable feature computation","Real-time pipeline with automatic stale signal invalidation"],"cta":"View repo"}]},{"id":"battleship","title":"Battleship","slug":"battleship","oneLine":"A browser-based Battleship game with AI opponent, drag-and-drop ship placement, and animated attack sequences.","thesis":"A classic grid-based strategy game implemented in vanilla JavaScript, demonstrating DOM manipulation, event-driven game state management, and a hunt-and-target AI opponent.","status":"shipped","domains":["Web Development","Game Development"],"roles":["Developer"],"stack":["JavaScript","HTML","CSS"],"skills":["typescript"],"featured":false,"order":15,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"bs-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/battleship","description":"Battleship source"},{"id":"bs-art-2","label":"Live Demo","kind":"demo","url":"https://danielboehmeiii.github.io/battleship","description":"Play in browser"}],"audienceSummaries":[{"mode":"recruiter","headline":"Battleship game with hunt-and-target AI in vanilla JavaScript","bullets":["Fully playable browser game with drag-and-drop ship placement","Hunt-and-target AI with directed search after first hit","Clean separation between game logic and DOM rendering"],"cta":"Play live"}]},{"id":"weather-app","title":"Weather App","slug":"weather-app","oneLine":"A weather dashboard with location search, geolocation default, current conditions, and a 5-day forecast.","thesis":"A weather dashboard built to practice async JavaScript, API integration, and responsive UI design — specifically graceful error handling and loading state management for network requests.","status":"shipped","domains":["Web Development","Data Visualization"],"roles":["Developer"],"stack":["JavaScript","HTML","CSS","OpenWeatherMap API"],"skills":["typescript"],"featured":false,"order":16,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"wa-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/weather-app","description":"Weather App source"},{"id":"wa-art-2","label":"Live Demo","kind":"demo","url":"https://danielboehmeiii.github.io/weather-app","description":"Live app"}],"audienceSummaries":[{"mode":"recruiter","headline":"Weather dashboard with geolocation, live API data, and 5-day forecast","bullets":["Fetches live weather data from OpenWeatherMap API with async/await","Geolocation default with city search fallback","Responsive design for mobile and desktop"],"cta":"View live"}]},{"id":"library","title":"Library","slug":"library","oneLine":"A personal book tracking app with CRUD operations, read-status tracking, and localStorage persistence.","thesis":"A library management app built to practice JavaScript OOP patterns, form validation, and local storage — demonstrating clean CRUD implementation and state management without a backend.","status":"shipped","domains":["Web Development"],"roles":["Developer"],"stack":["JavaScript","HTML","CSS"],"skills":["typescript"],"featured":false,"order":17,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"lib-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Library","description":"Library source"},{"id":"lib-art-2","label":"Live Demo","kind":"demo","url":"https://danielboehmeiii.github.io/Library","description":"Live app"}],"audienceSummaries":[{"mode":"recruiter","headline":"Book tracking app with OOP patterns and localStorage persistence","bullets":["CRUD operations for book management with form validation","localStorage persistence with JSON serialization","Vanilla JavaScript OOP with Book class and Library singleton"],"cta":"View live"}]},{"id":"todo-list","title":"To-Do List","slug":"todo-list","oneLine":"A task manager with project organization, priority levels, due dates, and localStorage persistence.","thesis":"A to-do list app built to practice DOM manipulation, event handling, date operations, and persistent client-side state — with project grouping and priority filtering as the core UX challenges.","status":"shipped","domains":["Web Development"],"roles":["Developer"],"stack":["JavaScript","HTML","CSS"],"skills":["typescript"],"featured":false,"order":18,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"todo-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/To-Do-List","description":"To-Do List source"},{"id":"todo-art-2","label":"Live Demo","kind":"demo","url":"https://danielboehmeiii.github.io/To-Do-List","description":"Live app"}],"audienceSummaries":[{"mode":"recruiter","headline":"Task manager with project organization, priorities, and localStorage","bullets":["Project-based task grouping with CRUD operations","Priority levels, due dates, and overdue visual indicators","localStorage persistence with full state serialization"],"cta":"View live"}]},{"id":"calculator","title":"Calculator","slug":"calculator","oneLine":"A browser calculator with keyboard support, operator chaining, and floating-point display rounding.","thesis":"A calculator implementation built to practice event handling, floating-point arithmetic edge cases, and display state management — specifically the subtle bugs in operator chaining and decimal rounding that naive implementations get wrong.","status":"shipped","domains":["Web Development"],"roles":["Developer"],"stack":["JavaScript","HTML","CSS"],"skills":["typescript"],"featured":false,"order":19,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"calc-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/Calculator","description":"Calculator source"},{"id":"calc-art-2","label":"Live Demo","kind":"demo","url":"https://danielboehmeiii.github.io/Calculator","description":"Live app"}],"audienceSummaries":[{"mode":"recruiter","headline":"Calculator with keyboard support and explicit state machine for edge case handling","bullets":["Explicit state machine handles operator chaining and consecutive operations correctly","Floating-point display rounding eliminates floating-point noise","Full keyboard support for digits, operators, Enter, Backspace, Escape"],"cta":"Try it live"}]},{"id":"etch-a-sketch","title":"Etch-A-Sketch","slug":"etch-a-sketch","oneLine":"A browser drawing tool with a resizable grid, color modes, and hover-activated pixel painting.","thesis":"An Etch-A-Sketch implementation built to practice dynamic DOM generation, event delegation at scale, and per-cell state management with CSS Grid.","status":"shipped","domains":["Web Development"],"roles":["Developer"],"stack":["JavaScript","HTML","CSS"],"skills":["typescript"],"featured":false,"order":20,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"eas-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/etch-a-sketch","description":"Etch-A-Sketch source"},{"id":"eas-art-2","label":"Live Demo","kind":"demo","url":"https://danielboehmeiii.github.io/etch-a-sketch","description":"Live app"}],"audienceSummaries":[{"mode":"recruiter","headline":"Etch-A-Sketch with CSS Grid, event delegation, and multiple color modes","bullets":["Dynamically generated N×N CSS Grid with single delegated listener (not per-cell)","Three color modes: single color, random RGB, progressive darkening","Configurable grid size up to 100×100 with clean regeneration"],"cta":"Try it live"}]},{"id":"rock-paper-scissors","title":"Rock Paper Scissors","slug":"rock-paper-scissors","oneLine":"A browser-based Rock Paper Scissors game with score tracking, round history, and animated result display.","thesis":"A Rock Paper Scissors game built as an introduction to JavaScript event handling, conditional logic, and DOM updates — the canonical first interactive web project.","status":"shipped","domains":["Web Development"],"roles":["Developer"],"stack":["JavaScript","HTML","CSS"],"skills":["typescript"],"featured":false,"order":21,"metrics":[],"claims":[],"evidence":[],"artifacts":[{"id":"rps-art-1","label":"GitHub Repo","kind":"repo","url":"https://github.com/DanielBoehmeIII/rock-paper-scissors","description":"Rock Paper Scissors source"},{"id":"rps-art-2","label":"Live Demo","kind":"demo","url":"https://danielboehmeiii.github.io/rock-paper-scissors","description":"Live app"}],"audienceSummaries":[{"mode":"recruiter","headline":"Rock Paper Scissors game with score tracking and round history","bullets":["Interactive game with random computer opponent and win/lose/draw detection","Score tracking and styled round history","Clean event handling and DOM state management"],"cta":"Play live"}]}],"meta":{"version":"1.0","generatedAt":"2026-11-19T09:10:22.535Z","count":21,"statuses":{"shipped":8,"prototype":12,"research":1,"concept":0}}}