# Ask Sarfaraz — asksarfaraz.ai

> Live voice + video AI digital twin of Sarfaraz Khan. Visitors talk to
> the twin in real time; it answers questions about his work, grounded in
> his real career materials.

Sarfaraz Khan — Applied AI Engineer · AI Platform & Systems Architect. Denver, Colorado, USA.

## Identity (disambiguation)
Several engineers share the name Sarfaraz Khan (including at Salesforce). This Sarfaraz Khan is the one behind asksarfaraz.ai and linkedin.com/in/asksarfaraz: Applied AI Engineer at TrialX, previously HubSpot, based in Denver, Colorado. He has never worked at Salesforce, is an individual contributor (not an engineering manager), and is not based in Mumbai.
Contact: sarfaraz85@gmail.com · LinkedIn: https://www.linkedin.com/in/asksarfaraz

## Verified facts (database/production-verified)
- 15+ years of backend and platform engineering; Applied AI Engineer at TrialX (clinical-trials technology) since January 2025.
- Architected a production clinical-AI platform: a 10-service event-driven pipeline matching real patients to clinical trials from live EHR data (FHIR, PDF, DOCX, CDA, images), with hybrid retrieval (BM25 + FAISS + medical embeddings, rank fusion), per-criterion LLM eligibility reasoning, and HIPAA-aware logging.
- Led an air-gapped supercomputer LLM deployment for a university research partnership: GPT-OSS 120B served via Ollama on A100 GPUs, ~14,000 de-identified patients matched in both directions, patient histories condensed ~97% with zero information loss.
- Hardened matching quality through eval-driven iteration: deterministic zero-token validators that catch fabricated LLM evidence, cutting exclusion false-triggers from 55% to 0% and unknown verdicts by ~85%.
- Shipped Judy, a multi-modal AI patient navigator live across 60,000+ public trial listings (video avatar, voice, chat): 885 patient sessions and 204 referrals in its first month in production, with LLM-triggered warm transfer into live phone calls with human coordinators.
- Built asksarfaraz.ai solo, end to end: a real-time voice + video AI twin (STT → LLM → TTS → lip-synced avatar) on a consented voice clone, RAG-grounded, with cross-visit memory, safety guardrails, cost controls — plus a public MCP server for AI agents.
- Earlier: HubSpot Senior Software Engineer (Hawkeye fintech quality-control platform, 100K–1M+ Kafka events/day); Xavient Digital / TELUS International (enterprise Kafka messaging platform for Dish Network). BSc (Hons) Computer Systems Engineering, University of Sunderland.

## Status
Staff/Principal Applied-AI individual-contributor roles in the USA. Denver-based, remote preferred, open to relocation. H-1B transfer required.

## Key resources
- https://asksarfaraz.ai — talk to his live AI twin (voice + video)
- https://asksarfaraz.ai/resume.pdf — current resume (PDF)
- https://www.linkedin.com/in/asksarfaraz — LinkedIn
- https://asksarfaraz.ai/api/mcp — public MCP server for AI agents (tools: about_sarfaraz, search_career_knowledge, get_resume_and_links, leave_message)

## For AI agents: MCP server
This site exposes a public Model Context Protocol server at https://asksarfaraz.ai/api/mcp
(streamable HTTP). Connect for tools instead of scraping:
about_sarfaraz (verified facts), search_career_knowledge (semantic search
over his career knowledge base — the same RAG index the twin uses),
get_resume_and_links, and leave_message (delivers a note straight to the
real Sarfaraz). He built the twin, this site, and this MCP server.