Projects

Realtime trace database and BM25 search MCP that turns Claude Code and Codex sessions into a searchable long-term memory layer.

• Realtime ingest
• BM25 search
• Agent memory

Open repository

Rule-based prose linter that scores formulaic AI writing patterns without an LLM judge, then returns concrete violations and revision advice.

• Prose linting
• CLI + MCP
• Rule engine

Open repository

Lightweight Zotero MCP server that lets agents search local libraries, browse collections, inspect items, and ingest papers by DOI or arXiv ID.

• Zotero
• Research workflow
• Paper ingestion

Open repository

PyTorch-inspired framework for composing, tracing, and optimizing LLM inference pipelines with automatic concurrency and resource management.

• LLM pipelines
• Execution DAGs
• Optimization

Open repository

Static, cited tracker of synthetic-token usage in open-weight LLM pretraining, with hoverable quotes tied directly back to technical reports and model cards.

• LLM tracking
• Cited data
• Static site

Open repository

General library for generating high-quality synthetic datasets from scratch or seed data, with validators, samplers, and LLM-backed generation workflows.

• Synthetic data
• Validation
• Seed or scratch

Open repository