Backend engineer working where reliability is the feature — FastAPI services, PostgreSQL at scale, Kubernetes operations, and observable LLM pipelines. I care about the unglamorous parts: latency, schema design, and tracing that makes 2 a.m. debugging survivable.
Currently shipping production backends at Flywheel for Sarlaben — Amul's multilingual AI chat + voice assistant for farmers.
role Backend Engineer · Flywheel
study B.Tech ICE · NSUT Delhi · 2023–present
focus APIs · data models · LLM telemetry · cluster ops
looking SDE / internship · 2026 · backend, full-stack, platform
- Unifying chat + voice into one FastAPI surface (≈40% less duplicate code)
- Owning Langfuse tracing across LLM workflows so failures are visible, not mysterious
- Running Kubernetes node lifecycle, autoscaling, and scheduling (cordon / drain / taints)
- Tuning PostgreSQL paths — 2,000+ glossary entries, ~30% lower query latency
Shipped on Amul AI's chat + voice stack — milk-collection lookups, vet / AI-technician booking, and a 5-turn non-meaningful call-termination gate that hangs up instead of looping the full LLM pipeline.
| signal | value |
|---|---|
| query latency | ~30% down |
| duplicate backend code | 40% removed |
| milk / payout lookups | 1,200+ in production |
| call bookings | 650+ in production |
| graceful hangups | 2,000+ (≈ $0.40–$0.50 saved per avoided loop) |
| glossary entries | 2,000+ with live updates, no redeploy |
| RBAC Banking | JWT + DB-backed permissions on FastAPI / PostgreSQL. Roles can grow without a schema rewrite; authz is revalidated on the request, not just baked into the token. | Python FastAPI PostgreSQL RBAC |
| Duolingo Clone | Full-stack learning path: lessons, XP / hearts / streaks, weekly leaderboard. Next.js frontend, FastAPI + Alembic backend. Live demo. | TypeScript Next.js FastAPI |
| amul-oan-api | OpenAgriNet stack for Amul — conversational chat APIs, farmer tools, moderation, translation pipelines. | Python LLM Langfuse |
More context, writeups, and the Objection / Orbit builds: yashgoel1331.github.io.
When LLM pipelines misbehave, I'd rather read the trace than guess.
Open to 2026 software-engineering internships and full-time roles — backend, full-stack, or platform. If you're hiring for systems that need to stay up, I'd like to hear about it.
portfolio · yashgoel31105@gmail.com · linkedin/in/yash-goel
uptime 100% · last deploy: now




