I build systems that survive real users —
from real-time infrastructure to AI pipelines to production backends.

I work at the intersection of product intent and systems design — building and stabilizing backends that run under real conditions, not just ideal ones.

Most of my work spans three areas:

• Real-time systems — event-driven backends, voice rooms, live data feeds, low-latency pipelines
• Applied AI — LangChain agents, multi-agent orchestration, LLM-integrated product features
• Production infrastructure — system stabilization, performance optimization, cost-aware architecture

I've mostly worked as the sole or primary engineer at early-stage startups, which means owning systems end-to-end — from architecture decisions to production deploys.

Led all technical systems across a multi-brand group as the sole engineer.

• Engineered a 14-phase automated AI SEO pipeline — transforming raw crawls into evidence-backed strategy playbooks using Crawl4AI, LiteLLM, and multi-agent orchestration. Reduced manual audit time from weeks to hours.
• Built a React enterprise website (Jayshree International) with atomic design system — achieved 0.4s LCP on an image-and-video-heavy page.
• Reduced LCP from 12s → 1.7s on Shopify storefronts through lazy loading, asset optimization, and delivery strategy.

Node.js React Agentic AI Crawl4AI LiteLLM Shopify (Liquid) Core Web Vitals

Built an AI-powered edtech platform entirely from scratch, end-to-end.

• Full-stack across 30+ screens in React/Redux + Node.js/MongoDB — auth, test engine, course delivery, AI tutoring. Load-tested to 1,000 concurrent users at <180ms API response.
• Built a LangChain + Google AI pipeline for syllabus-scoped Q&A, notes, flashcards, and interactive voice Q&A. Validated outputs against a structured test suite for accuracy, edge cases, and safety.
• Replaced Google Analytics with a custom behavioral analytics pipeline — streaks, watch time, quiz performance — with IndexedDB buffering and batched sync.
• Built a live class system replacing paid Zoom integration — instructor video prioritization, teacher controls, live polls.

Node.js React Redux MongoDB LangChain Google AI AWS WebSockets IndexedDB

Inherited a broken production backend for a social voice app; stabilized and production-readied it solo in one month.

• Took ownership of a fragmented, undocumented Node.js codebase — refactored into OOP-based, feature-scoped modules while preserving all production data and behavior.
• Stabilized real-time voice rooms (seat-based speaking, gifting, locking, listener-only modes) using event-driven architecture. Surface and resolved race conditions under load-testing.
• Built hierarchical monetization across a 4-tier principal hierarchy — wallet flows, commissions, salaries, moderation, bans. Backed by automated end-to-end tests.

Node.js MongoDB Event-Driven Architecture WebSockets Agenda AWS

Live market data pipeline for a proprietary NSE futures trading system.

• Designed a Python WebSocket-to-TCP bridge feeding live NSE futures data into a MATLAB execution environment — keeping the trading system's interface unchanged.
• Built MessagePack binary serialization + auto-reconnection for feed resilience. Implemented in-memory O(1) snapshot access for sub-millisecond reads.
• Validated across millions of market ticks in live conditions with zero data loss.

Python MATLAB WebSockets TCP/IP MessagePack Parallel Computing

 

B.Tech Computer Science & Engineering
Cooch Behar Government Engineering College · MAKAUT · 2020 – 2024 · CGPA: 8.75