Open-source tooling built for real scholarship — each one in production use on my
own pipelines before it was ever released.
AI research agent with cryptographic citation verification
A hybrid Rust + Python system that reads research materials, indexes them into a Merkle-verified knowledge base, and answers questions with deterministic, trace-bound citations: every claim maps to a specific text passage verified by SHA-256 with a full Merkle proof. Seven deterministic agents form the pipeline — ingestion, indexing, query planning, BM25 retrieval, clarification, generation, integrity verification — with a Rust (ratatui) TUI for interactive sessions. Published on PyPI as cite-extractor.
Structured citation extraction from academic sources
Python library that extracts structured CSL-JSON citation data from PDFs (digital and scanned), URLs, and media files. Multilingual OCR including vertical CJK, document-type classification (journal, book, thesis, chapter), and multiple citation styles (Chicago, APA, MLA). Pluggable LLM backends: Ollama, Gemini, OpenAI. Grounding generation in verified source citations reduces LLM hallucination in scholarly contexts.
Translation Assistant of Church Fathers — multilingual morphology platform
Full-stack application (FastAPI + Tauri + Keycloak OIDC) for Greek, Syriac, Latin, and Coptic morphological analysis and lemma lookup. SQLite-backed accentless Greek morphology index, real-time dictionary cards drawing on LSJ, Payne Smith, Lampe, and Logeion, LLM fallback for disambiguation, document upload with OCR decision routing, and Redis rate limiting behind a Caddy reverse proxy.
Multilingual manuscript OCR/HTR toolkit
Route-aware OCR/HTR engine integrating Kraken and Tesseract for Greek, Latin, Syriac (Estrangela, Serto, East), Coptic, Armenian, and Geʽez. CER/WER benchmark reporting, Kraken model-training pipeline, FastAPI service APIs, Gradio demo UI, and RunPod training-job submission. The primary Latin print model (CATMuS-Print Large) reaches 98.56% accuracy at CER 1.44%; handwritten support covers CATMuS Medieval, 8th–15th-century manuscripts.
Multimedia → structured Markdown with AI rewriting
CLI tool converting video, audio, and URLs into VTT/Markdown, then rewriting or translating: transcription → academic rewrite, bilingual translation (DeepL first, LLM fallback), PPT slide-plus-speech combination, and batch directory processing. Configurable via CLI, Python API, or YAML. Runs in production as the transcription layer of an 80+ lecture video pipeline. Published on PyPI.
Video editing CLI with YouTube subtitles and upload
CLI for video segment extraction, text overlays, and color/gamma/speed effects, plus a subtitle pipeline — YouTube transcript download (original and translated) → SRT conversion → MoviePy overlay rendering — and upload with multi-account credential management, playlist assignment, and thumbnail extraction. Multiple LLM providers for intelligent subtitle translation.
JobFinder
In production use — source available on request jobfinder
Agent-native job-search system
An agent-native system for academic and Digital Humanities job searches: deterministic Python plumbing (CV ingestion, job-source search and normalization, SQLite storage, regex ATS and legitimacy signals, application tracking) while the agent supplies reasoning through portable SKILL.md skills. No embedded LLM; outputs are drafts and checklists, and every external action requires human approval.
PythonTyperSQLAlchemySQLitePydantic