Oncology Trial Matcher — Ammaad Tehseen

Closed — Private source — described by architecture only2026

Full-stack developer · Hashlogics

Three readings

Client: Matches cancer patients to the clinical trials that actually fit them, then writes the report a clinician, nurse or insurer needs — with the compliance a healthcare product has to carry.
Recruiter: Full-stack on a HIPAA-grade oncology platform: NestJS, Prisma and Postgres behind a React and TypeScript front end, with BullMQ report queues, Socket.IO, S3 and Stripe. Also ran the platform's compliance programme in Sprinto.
Engineer: ClinicalTrials.gov returns anything that mentions the words, so a custom ranker re-scores the results on token-level cancer, biomarker and gene-symbol matches and the patient-relevant trials surface first.

What it does

Patients matched to open trials by cancer type, stage, biomarkers and line of therapy
AI reports with NCCN guideline summaries and drug-efficacy context, delivered as PDFs
Separate portals for patients, nurses, insurers and admins — including bulk report runs for insurers
Compliance handled end to end: audit logs, two-factor auth, encrypted email and a Sprinto-run programme

Relevance ranking over ClinicalTrials.gov v2 at token level, so “lung” matches “non-small cell lung cancer” but not “plunge”
Report generation runs on a BullMQ queue, with progress pushed to the browser over Socket.IO
Every LLM call goes through a circuit breaker, streams where it can and logs its token cost; system prompts are versioned in the database
De-identified patient profiles kept apart from identified ones, and every sensitive action written to an audit log
TOTP two-factor auth, HIPAA-compliant email through Paubox, presigned S3 uploads, helmet and rate limiting on the API

source private — architecture only

Source is private. Happy to walk through the architecture on a call — ammadtehseenkhan@gmail.com