Full-stack AI product engineer, 2 years at Addie, an AI counseling platform for 2,000+ students. Our engineering team is me and one other engineer, so I own features from data model to deploy: 560+ merged PRs, project lead on 11 projects.
Some of the work:
- Multi-channel AI conversations: guided conversations and multi-step journeys over web, SMS, WhatsApp and voice (13,000+ assigned; 2,200+ student voice calls, 370+ hours). I led browser voice on OpenAI Realtime and moved it to a real WebRTC connection, which took fake turns per call on speakerphone to zero.
- Counselor retrieval agent (LangGraph, 9 tools, server-bound data scope, hard limits on rounds and time). A planted-fact recall eval caught a regression our demos hid; the redesign lifted recall from 3% to 43% at a third of the tokens.
- Meeting transcription: picked AssemblyAI after a vendor test (99.4% accuracy, half the cost), built the consent flow state recording laws require. Live at 5 schools.
- Found our deploys reporting success while production ran 3-day-old code; rebuilt the pipeline with verification, real rollback and alerts.
Side project: OfferRing, a Chrome extension and Next.js app that ranks new postings, fills applications without ever submitting, and drafts answers only from facts I wrote down. Every AI feature has its own eval and model.
Looking for: full-stack or AI product engineer roles on small teams with real ownership. Full-time or contract.
Remote: Yes. Canadian permanent resident; for US companies, contractor or EOR
Willing to relocate: No (happy to travel)
Technologies: TypeScript, React, Next.js, Node.js, Python, FastAPI, Celery, PostgreSQL, Prisma, Redis, Qdrant, LangGraph, OpenAI Realtime, WebRTC, Twilio, AssemblyAI, Claude API, Docker, GCP, GitHub Actions
Résumé/CV: https://dempsey-portfolio.vercel.app/ (resume and recorded demos)
GitHub: https://github.com/zsg0428
Email: zsg258852 [at] gmail [dot] com
Full-stack AI product engineer, 2 years at Addie, an AI counseling platform for 2,000+ students. Our engineering team is me and one other engineer, so I own features from data model to deploy: 560+ merged PRs, project lead on 11 projects.
Some of the work:
- Multi-channel AI conversations: guided conversations and multi-step journeys over web, SMS, WhatsApp and voice (13,000+ assigned; 2,200+ student voice calls, 370+ hours). I led browser voice on OpenAI Realtime and moved it to a real WebRTC connection, which took fake turns per call on speakerphone to zero.
- Counselor retrieval agent (LangGraph, 9 tools, server-bound data scope, hard limits on rounds and time). A planted-fact recall eval caught a regression our demos hid; the redesign lifted recall from 3% to 43% at a third of the tokens.
- Meeting transcription: picked AssemblyAI after a vendor test (99.4% accuracy, half the cost), built the consent flow state recording laws require. Live at 5 schools.
- Found our deploys reporting success while production ran 3-day-old code; rebuilt the pipeline with verification, real rollback and alerts.
Side project: OfferRing, a Chrome extension and Next.js app that ranks new postings, fills applications without ever submitting, and drafts answers only from facts I wrote down. Every AI feature has its own eval and model.
Looking for: full-stack or AI product engineer roles on small teams with real ownership. Full-time or contract.