# Nayab Rehmat

> Software engineer who has built agentic AI systems in production healthcare.

- Site: https://nayabdev.com
- Previously Software Engineer at Inception Fertility (2023 — 2026)
- Location: Austin, TX. Open to SF or NYC.
- Education: Texas A&M University, B.S. in Computer Science and Engineering

## Products

- [Ambora](https://ambora.io): Code with your hands in your pockets. Describe a change by voice or text from your phone; Ambora plans the work, explains its approach, then writes and tests it on a branch so it's waiting when you sit back down. Auth, payments, and config changes sit behind explicit approval gates, and nothing reaches GitHub without you saying so.
- [Backlog](https://backlog-hazel-beta.vercel.app): Your reading, back as a morning letter. Share an article from any app and it disappears into a queue; at 6am you get one edited five-minute email covering everything you saved, with editorial notes on what's worth your time and links straight to the originals. No feed, no recommendations, the only thing in it is what you chose.
- [VCDB](https://vcdb-nine.vercel.app): A venture-firm directory built for job hunting, sitting on top of the pipeline that fills it in. I imported 1,350 firms as names alone and let a worker do the rest: it resolves each firm's domain from Brave search results, scores the match on name-token overlap and venture context, then probes the site for portfolio, careers, and jobs-board URLs and tags the platform behind them, Getro, Ashby, Greenhouse, or Lever. Anything under a 0.7 confidence gate goes to a review queue instead of the directory. The worker claims batches of 100 from a retrying job queue in Postgres on a ten-minute GitHub Actions cron, so the whole backfill ran unattended and new firms enrich themselves. The Next.js front end reads the table with the anon key only, filtered, sorted, and paginated straight from the URL.
- [GridScope](https://nayabdev.com/#built): A grid-intelligence platform that fuses three government data sources, the ERCOT interconnection queue, TCEQ air permits, and PUCT dockets, into one de-duplicated, geolocated view of 1,100+ Texas projects across ~12.7k historical snapshots. I led the backend. None of the three sources share an ID or a coordinate, so I built coordinate-free entity resolution using token-set Jaccard matching on name and county, tiered into resolved, review, and unresolved, plus two-tier geolocation from Census county centroids to exact EPA FRS coordinates so every project still got a map pin. I owned ingestion end to end: multi-sheet ERCOT .xlsx reports, the TCEQ Socrata API, and PUCT docket filings, normalized into snapshotted Postgres tables.
- [CRTX](https://crtx.chat): A full-stack RAG platform for semantic search over your own documents. Upload PDFs and URLs; CRTX chunks, deduplicates, and embeds them for question-answering over a private corpus. Supabase Auth, JWT-gated routes, and share links that keep ingest and query permissions separate. I piloted it with a client, mapped their document-review workflow, and shipped features straight from their feedback.

## Skill packs

Downloadable Claude Code skills. Machine-readable index: https://nayabdev.com/skills/index.json

- [visual-design-loop](https://nayabdev.com/skills/visual-design-loop.zip): /visual-design-loop — A build-and-refine process for web UI that keeps it from looking AI-generated. It closes the three gaps that usually cause it: pin down a real brief before writing code, lock a token system before writing CSS, then iterate against actual screenshots instead of trusting the markup. Ships with a seven-pass critique rubric, a guide to turning inspiration images into a transferable brief, and a Playwright capture script that audits every pass for overflow, contrast failures, small tap targets, token sprawl, and console errors.
- [plain-outbound-writing](https://nayabdev.com/skills/plain-outbound-writing.zip): /plain-outbound-writing — A final pass for anything a person will read as a message from you: email, Slack, DMs, cover letters, outreach. Most AI-sounding writing isn't wrong, it's performed. It keeps signaling that a point was made instead of just making it. This names each move that produces the effect, from the "that's not X, that's Y" reveal to stacked sentence fragments to image pairs that stand in for the actual instruction, and gives the plain replacement for each.

## Pages

- [Home](https://nayabdev.com/)
- [Home (Markdown)](https://nayabdev.com/index.md)
- [About (Markdown)](https://nayabdev.com/about.md)
- [Ambora (Markdown)](https://nayabdev.com/projects/ambora.md)
- [Backlog (Markdown)](https://nayabdev.com/projects/backlog.md)
- [VCDB (Markdown)](https://nayabdev.com/projects/vcdb.md)
- [GridScope (Markdown)](https://nayabdev.com/projects/gridscope.md)
- [CRTX (Markdown)](https://nayabdev.com/projects/crtx.md)
- [visual-design-loop (HTML)](https://nayabdev.com/skills/visual-design-loop/)
- [plain-outbound-writing (HTML)](https://nayabdev.com/skills/plain-outbound-writing/)
- [visual-design-loop (Markdown)](https://nayabdev.com/skills/visual-design-loop.md)
- [plain-outbound-writing (Markdown)](https://nayabdev.com/skills/plain-outbound-writing.md)
- Full text: https://nayabdev.com/llms-full.txt

## Links

- [GitHub](https://github.com/mnr282001)
- [LinkedIn](https://www.linkedin.com/in/nayab-rehmat/)
- [X](https://x.com/m_nayabrehmat)
- Email: nayabrehmat28@gmail.com