AI systems · reproducible research · practical tooling
I turn technical ideas into small, auditable, and runnable systems.
- Knowledge-graph-guided data and task synthesis
- Reliable post-training pipelines for language models
- Retrieval, evaluation, and verifier-oriented workflows
- Reproducible implementations of ideas from technical reports
|
A reproducible Python pipeline that grows a knowledge graph, retrieves public evidence, synthesizes grounded post-training tasks, and validates the outputs.
|
Python · CLI tooling · REST APIs · Knowledge graphs · Retrieval · LLM evaluation · Git
- Reproducible by default — explicit configs, deterministic paths, and resumable runs.
- Evidence before claims — grounded inputs, clear boundaries, and honest limitations.
- Small pieces, strong interfaces — readable modules that are easy to test and replace.
Building in public, one useful system at a time.