FullStack and platform engineer — Go, Java, PostgreSQL, Angular. I build and run Tracktions solo, and work on internal AI developer tooling at BNY.
A trading journal platform for Indian retail traders on NSE/BSE. I built the whole thing: Go/Gin API on PostgreSQL, Angular 17 frontend, deployed on AWS.
Some decisions I'd defend:
- Server-sent events, not WebSockets. Trade imports are long-running jobs and progress is strictly server-to-client. SSE gave me streaming progress over plain HTTP with no extra protocol surface to secure or scale.
- Dropped Redis for an in-memory job queue. Redis was doing job queuing and little else — one more stateful service to run, monitor and pay for. Moving the queue in-process removed a whole moving part with no loss in throughput at current volume, and it's a decision I know how to reverse when it stops holding.
- Direct broker sync over manual entry. Zerodha Kite integration plus CSV/Excel import, because the fastest way to kill a journaling habit is to make people type their trades in twice.
Also: Razorpay subscription billing, AWS SES/S3/CloudWatch, JWT cookie sessions, 35 schema migrations, ~350 Go files. Closed source — happy to walk through any of it.
Internal AI developer tooling — an air-gapped marketplace for configurable AI skills and MCP servers, and a multi-agent LLM pipeline that automates UAT across our ODS datasets and microservices. Enterprise code, so none of it is public.
tracktions-architecture — the decisions behind the system above, each with the condition that would reverse it. Why SSE and not WebSockets, why Redis got deleted rather than grown, where the design runs out of room.
codebase-context-pilot — I built a code-graph context system to make coding agents cheaper on a 354-file Go codebase, wrote the pass/fail rule before collecting any data, measured it across 10 tasks, and shut it down when it missed. The repo is the full record: methodology, numbers, the two measurement bugs that invalidated the first pass, and what I'd try next. Verdict was no-go; the interesting parts are why, and the one task shape where it won decisively.
Competitive-Coding-From-scratch — competitive programming resources and solutions collected while learning. Still the most useful thing I've published, judging by the stars.
Go · Java / Spring Boot · TypeScript · PostgreSQL · Angular · AWS · Docker · CI/CD LLM orchestration · multi-agent systems · MCP