Yash Goel builds systems that stay up.
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.
Profile
B.Tech ICE · Netaji Subhas University of Technology · Delhi · 2023–presentI'm a backend-leaning full-stack engineer who treats correctness and observability as first-class features, not afterthoughts.
Most of my work lives below the UI: consolidating fragmented services into clean APIs, tuning database schemas so queries stay fast as data grows, and wiring telemetry through AI workflows so failures are visible instead of mysterious.
At Flywheel I shipped production backends end to end: unified chat and voice in one FastAPI service, optimized a scaling glossary for latency, and instrumented LLM pipelines across the stack. I operated Kubernetes clusters directly—node lifecycle, autoscaling, scheduling—and led Pedal Playground’s technical track, a pan-India analog pedal build event that ties my software work back to where signal and systems start.
Experience
// production work, in chronological orderPRESENT
Backend Engineer
Flywheel- Engineered backend systems for Sarlaben, Amul's multilingual AI chat and voice assistant serving farmers, unifying chat and voice services into a single FastAPI application and reducing code redundancy by 40%.
- Shipped milk-collection, vet and AI-technician booking tools on Amul AI's chat + voice stack, enabling farmers to check payout/deduction history and schedule livestock services in Gujarati—1,200+ lookups and 650+ call bookings in production, replacing manual passbook reconciliation and phone bookings.
- Designed and shipped a 5-turn non-meaningful call-termination gate on Amul AIs voice helpline, automatically ending calls stuck onunclear speech, STT failures, or repeated fillers instead of looping the full LLM pipelinetriggering 500+ graceful hangups, saving anestimated $0.40-$0.50 per call in avoided agent, STT/TTS, and telephony spend.
- Owned Langfuse telemetry across chat and voice AI backends, lifting debugging and observability efficiency ~40% through centralized tracing of LLM workflows.
- Ran Kubernetes cluster operations — node lifecycle, autoscaling validation, and scheduling with cordon, drain, taints, and node selectors.
- Reworked the PostgreSQL schema and query paths for 2,000+ glossary entries — ~30% lower latency and real-time updates without redeployment.
TRACK
Technical Lead — Pedal Playground
NSUT · Techno-Music Build Event- Led the technical track for Pedal Playground, a pan-India event where 15+ college teams designed and built their own guitar effect pedals from scratch.
- Set the circuitry brief and judged builds as teams tuned resistance and capacitance across analog stages to shape gain, tone, and signal response.
- Bridged my Instrumentation & Control background with hands-on hardware — the same signal-path thinking I bring to backend pipelines.
Selected Builds
// self-directed systems, deployed & documentedObjection
active buildDesigned the foundation around structured courtroom logic before adding LLM agents. Built Pydantic case-file models for charges, witnesses, evidence, timelines, contradictions, and hidden/public visibility. Added FastAPI endpoints and tests to ensure player-safe case data does not leak hidden facts.
RBAC Banking Backend
A role-based access control system governing dashboard and API permissions across multiple user types, with secure financial-record APIs and authorization enforced at the application layer.
OpenAgriNet APIs
Contributions across the OpenAgriNet stack — voice and data APIs for agricultural use cases (incl. an AMUL implementation), working with conversational interfaces and structured data config.
Orbit
A responsive full-stack app: 35+ reusable, accessible, mobile-first UI components on the front, wired to a Node.js + MongoDB backend with clean state management and reliable data display.
Stack & Foundations
// what I reach for, grouped by layerLanguages
- Python
- C++
- SQL
- JavaScript
- HTML / CSS
Backend & Data
- FastAPI
- REST APIs
- Redis
- PostgreSQL
- MongoDB / MySQL
- Data Modeling
- Marqo
AI / LLM & Infra
- LLM Integration
- Langfuse Tracing
- Context Management
- Moderation Pipelines
- Kubernetes
- Linux · Git
Frontend & CS Core
- React.js / Next.js
- Tailwind CSS
- Data Structures & Algos
- DBMS
- OOP
Reliability under load
Payments and consulting platforms run on systems that can't fail quietly. Schema discipline, latency tuning, and tracing are exactly the muscles I've been building.
Services, not scripts
I think in services and contracts — consolidating backends, designing extensible permission models, and shipping APIs other teams can build on without breaking.
Observable LLM workflows
I've owned the telemetry layer for production AI backends. When LLM pipelines misbehave, I'd rather read the trace than guess.
Let's build
something reliable.↗
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.