AI Developer at NeLabs AIFounder of SearchInk
AI systems that answer from evidence.
I build the retrieval, permission and ranking layers that decide whether an AI answer can be trusted. Currently shipping SearchInk — private search over a company's own knowledge — and AI automation at NeLabs AI.
focus=["retrieval", "permissions", "ranking"]
SearchInk — private search over a company's own knowledge
AI Information Context Engine
SearchInk indexes the documents, mail and conversations a business already owns, then answers questions over them with citations instead of guesses. I own the retrieval and permission layer: scoring documents, routing the response by intent, and refusing to answer when the corpus does not support it.
- 1,022 docs
- 8 sources
- 1,207 permissions
- 0.4–0.7 s
Turning a full inbox into a ranked task queue
NeLabs AI Task Management System
Tasks arrive as email, WhatsApp, calls, calendar invites and documents. The system extracts actionable items with rules first, falls back to Llama 3.1 Nemotron only when confidence is low, then ranks every task with a seven-factor priority model.
- 5 channels
- 7-factor ranking
- PostgreSQL + pgvector
An assistant that runs on a real machine, every day
NeLabs AI Personal Assistant
A personal assistant wired into Gmail, Calendar and a private document corpus that runs continuously on a Linux server rather than in a notebook — OAuth2, MCP connectors, Caddy, PM2 and Postgres underneath.
- Gmail + Calendar
- Always-on VPS
- Postgres + pgvector
Building something that has to be trusted?
Open to work on retrieval, applied AI and the systems that keep an answer honest. Email is the fastest route.