azterizm
Remote: Yes
Willing to relocate: No
Technologies: Python, TypeScript, Node.js, SGLang, PyTorch, OpenAI, Anthropic and Gemini APIs, RAG, Remix, Prisma, Redis, SQLite
Résumé/CV: https://azterizm.com/Abdullah_Memon_CV.pdf
Email: [email protected]
My verifier catches wrong statements in LLM answers as they are written. It checks each sentence against its source and rewrites the ones that fail. On UK law questions it cut answers containing a wrong statement by up to 86%, while still answering at least 93% of questions: https://github.com/azterizm/legal-rag-verifier
A router sends each legal question to the exact law it cites, before any search runs and without a model call. Across 1,897 real citations it never picked the wrong law, and declined the 4% it could not place: https://github.com/azterizm/legal-rag-router
In October I reported two scoring bugs in Mastra's evals with a reproduction, and their CTO fixed both within nine hours: https://github.com/mastra-ai/mastra/issues/28736 . A fix I wrote for keyword scoring in the RAG search of Tether's QVAC is open as a PR: https://github.com/tetherto/qvac/pull/5467
Since 2021 I've built client software in TypeScript on Node.js. The roles I want are full-time and remote, in LLM evaluation, verification, RAG and agent infrastructure.
although trading disk space for sub second search is good choice. overall great technical execution!
That is a seriously good idea.
I tested Jev against local models. Confirmed that local architecture outperforms the closed cloud one marginally.
Limits mostly come from the cloud model. We are talking 712ms P50 latency versus 0.14ms locally on intent routing. Scoring costs linearly with volume. Cloud got its own limits.
An open source Jev would resolve all this immediately. Network, billing, RPM ceiling, all resolved. Additionally, we can do fine tuning, scale with hardware and freeze weights.