Soham S.

I'm an AI-first backend engineer. I build the APIs and the agent systems on top of them, in Node.js, TypeScript, and Go. What I work on: - AI agents for support work: specialist agents, a small LLM router into a stronger reasoning model, and RAG over versioned policies with citations - Guardrails around models: a deterministic policy engine so a prompt cannot move money, idempotent actions, and Postgres leases so two agents cannot act on one case - REST APIs in Node.js and Express for concurrent production workloads - Multi-tenant backends, including FHIR-compliant healthcare data models with tenant isolation - Performance: Redis caching, schema work, and indexing (cut one API's p95 latency from 2s to 200ms) - Real-time features with Redis Pub/Sub (sub-200ms follower updates and reactions) - Async pipelines and background jobs, including chunked audio ingestion and speech analysis - Payments: Stripe subscriptions with verified, retry-safe, idempotent webhooks - Observability with Prometheus and Grafana, plus Docker and Kubernetes Selected work: - Munshi, an AI agent for payment support. It resolves routine tickets and escalates to a human only when needed. Stack: Node.js, PostgreSQL with pgvector, Redis, BullMQ, and React. - Weavo, a real-time collaboration SDK. Implemented the YATA CRDT from scratch and published @weavo/client on npm (890+ weekly downloads), with cursor reconciliation, automatic reconnect, and offline buffering. - A Layer-7 HTTP load balancer in Go with round-robin routing, active health checks, retries, and failover. Stress-tested with 8.2M+ requests using k6. I write clean code, communicate early, and flag problems before deadlines slip. Send a short description of what you're building and I'll tell you how I'd approach it.