Software Development Engineer II·Metaforms AI

Prateek Goyal

Bengaluru, Karnataka, India

Product and platform engineer — I care most about AI systems that stay reliable, observable, and affordable once they leave the demo.

Now buildingDebrix

About

I am especially interested in AI systems and scaling them responsibly — not just calling a model, but designing workflows that degrade gracefully, stay within budget, and stay debuggable when things go wrong in the wild.

On the software side I have strong hands-on experience with agentic workflows: tool use, multi-step pipelines, guardrails, and human-in-the-loop patterns — aligned with where the field is heading as products mature (measurement and evals, routing and cost control, composable tool interfaces, and operational rigor instead of benchmark chasing).

I am tech-agnostic about the stack: I pick languages, runtimes, and cloud primitives for the constraint — not the logo. Lately that has meant backend services, queues and workers, data stores, observability, sandboxed execution for untrusted code, sync engines (optimistic updates, conflict handling, and keeping client and server honest), and real-time / streaming UX.

Experience

Metaforms AI
Software Development Engineer II · Bengaluru
May 2025 – Present
- Backend for a customer-facing Survey Programming Copilot; LLM-assisted workflows with guardrails (FastAPI, Tauri, React + Rust).
- Sandboxed Python execution with isolation and audit logging — safer untrusted code without host FS/network access.
- Excel-like grid: 10k+ cells, virtualization, undo/redo, Excel paste, optimistic sync with FastAPI.
- Owned releases end-to-end: API contracts, Sentry error budgets, fewer incidents via release gates and rollback automation.
Metaforms AI
Software Development Engineer I · Bengaluru
May 2024 – Apr 2025
- Migrated Voice Forms streaming to LiveKit (WebRTC) — lower latency and better UX on weak networks.
- Async job pipeline with Redis and worker pools for high-volume form submissions; Docker + blue-green on AWS ECS.
- Frontend performance: service worker caching, virtualized long lists — faster loads and smoother scrolling.

What I build with most

Skills

Backend systems, cloud infrastructure, AI workflows, and frontend delivery tools I reach for most often.

Core

FastAPI
Node.js
REST APIs
React
TypeScript
PostgreSQL
MongoDB

Cloud & tools

AWS (ECS, Lambda, S3, CloudFront, ALB, SQS, ECR, CloudWatch)
Docker
Redis
CI/CD
Firebase
Tauri
Git

Engineering

System design
Distributed systems
API design
Observability
Performance
LLM integration
Agentic workflows
Guardrails & eval mindset

Languages

TypeScript / JavaScript
Python
SQL
Rust

Education

Jabalpur Engineering College

B.Tech, Computer Science and Engineering

Nov 2020 – Jun 2024

CGPA 8.43 / 10

What I'm building

Independent work

Products I take from the first sketch through the operational details.

Founder & builder Controlled early access

Debrix

The interactive debugger for AI agents

debrix.io

Open a failed run. Hold the evidence constant. Test one change at a time. Then verify the real fix with no overrides and save it as a regression.

Local-first
IDE-native
Evidence-first

Small experiments

Small, opinionated desktop experiments — the kind of software I build for fun, to learn a stack end-to-end, or to scratch a very specific itch.

Slap My Mac
Menu bar app for Apple Silicon MacBooks
Tiny Tauri + React experiment: the laptop IMU listens for a sharp tap beside the trackpad, plays a sound, and increments a counter — with sensitivity, cooldown, optional custom audio, and a quick volume bump while it plays.
Tauri
React
TypeScript
Rust
GitHub
Apple Silicon for the motion sensor; other Macs can still use test slap and audio settings.

Coding profiles & highlights

Competitive programming profiles and selected contest results.

Achievements

Led team to the Smart India Hackathon 2022 Finals, securing a top 5 position among 125 teams. Certificate
Achieved regionalist in ICPC Amritapuri and qualified ICPC Kanpur preliminary round with a rank of 620.
Specialist at Codeforces with highest rating of 1470 and Knight at LeetCode with 2000+ rating.
Ranked 292 among 29k+ participants globally and AIR 69 in LeetCode Biweekly Contest 111. Post