backend engineer / product infrastructure
Backend engineer building product infrastructure, AI workflows, and developer tools.
current work
Early-stage CRM/marketing SaaS. I work across the backend pieces that make the product behave reliably: queues, APIs, read models, AI signals, telephony credits, permissions, and imports.
owned / queues, read models, AI signals, telephony credits, RBAC, imports
The lead page looked simple, but every filter hit a join-heavy query and the DB pool started struggling under modest load.
Moved the read path from 4-table joins to a materialized table updated by post-commit events.
Scheduled notifications and email campaigns needed delayed delivery without depending on one global scheduler.
Replaced a naive scheduler with RabbitMQ TTL + DLX delay queues.
Sales signals were scattered across transcripts, emails, notes, and CRM fields.
Built a queued signal pipeline that combines LLM tags with deterministic scoring.
The CRM needed audit trails for calls, emails, updates, and lead changes without making product actions feel slower.
Moved activity logging behind domain events so request paths did not wait on timeline writes.
Filtering and pipeline behavior had started spreading across CRM views as duplicated logic and hard-coded statuses.
Centralized filter definitions and moved lead/opportunity stages into a DB-backed pipeline model.
built outside work
These are not stack demos. Each one started from a specific failure mode or repeated engineering problem I wanted to understand better.
AI knowledge engine over codebases (Go)
The hard part was not calling an LLM. It was preserving code meaning before retrieval.
builtAST-aware chunking, hybrid retrieval, reranking, and a CLI that indexes a repo and answers questions end to end.
nexus index <path> nexus ask "where is auth handled?"
Distributed job orchestration (Go + Node.js)
Built after seeing how easy it is for background work to fail silently under real production conditions.
builtA job orchestration runtime for immediate, delayed, and cron jobs with agent execution, state tracking, heartbeats, and retry behavior.
CREATED -> RUNNING -> SUCCESS / FAILED / TIMEOUT
LLM reverse proxy (Go)
The same retry, routing, caching, and cost-tracking code kept appearing across projects.
builtA drop-in reverse proxy for OpenAI-compatible endpoints with provider routing, semantic caching, streaming, retries, metrics, and per-model cost tracking.
request -> route model -> cache check -> provider -> stream response
also shipped
technical notes
tools I reach for
dailyGo / TypeScript / Node.js / PostgreSQL / MySQL / Redis / RabbitMQ / BullMQ / Docker / AWS EC2
AI / LLMAzure OpenAI / OpenAI API / pgvector / embeddings / RAG pipelines
also usedVue.js / Prisma / Sequelize / Plivo (VoIP) / WhatsApp Business API / Razorpay
education
B.Tech, Computer ScienceVivekananda Global University / May 2021 – May 2025
contact