I'm a Software Engineer with 2 years of production experience at Tata Consultancy Services, currently pursuing an M.Tech in Computer Science at NIT Calicut.
I build high-performance backend systems for a living, dig into POSIX system calls for fun, and research Vision-Language Models for my thesis. I enjoy working at every layer — Java microservices, TCP sockets in C++, React frontends, and from-scratch neural networks.
- 🎓 M.Tech CS @ NIT Calicut (Expected 2027) · B.Tech IT @ Techno Main Salt Lake, Kolkata
- 💼 Ex-TCS — 2 years of enterprise Java/Spring Boot backend development
- ⚙️ Systems programmer — raw TCP sockets,
fork/exec,dup2, multithreading in C++ - 🔬 Researcher — Multimodal Anomaly Detection with Vision-Language Models
- 🤖 AI practitioner — LLM agent systems, from-scratch neural networks in NumPy, multi-agent simulations
- 📍 Calicut, Kerala · Open to remote & on-site opportunities
timeline
title From Kolkata to Calicut — via Production Systems
2019 : Started B.Tech (IT) at Techno Main Salt Lake, Kolkata
2023 : Graduated B.Tech : Joined TCS Bengaluru as Software Engineer
2025 : Completed 2 years at TCS : Started M.Tech (CS) at NIT Calicut
2026 : 88.5% AUROC zero-shot anomaly detection research : Built NegotiAgent — LLM agents + ABT protocol : LitRAG — evaluation-first RAG, 0.957 recall@1
2027 : M.Tech graduation (expected) : Open to SWE and research roles
|
Multimodal Anomaly Detection with Vision-Language Models Zero-shot & few-shot defect detection with CLIP-based VLMs on the MVTec AD benchmark — combining visual and language grounding for pixel-level defect localization. 📈 88.5% AUROC zero-shot on MVTec AD — no training on defect data
|
Evaluation-First RAG over Research Papers Retrieval-augmented QA with inline citations, built eval-first: a curated 52-question set, recall@k and faithfulness scoring, and a 30-config ablation of chunking, embeddings and reranking. Plain Python — no framework magic. 📊 0.957 recall@1 · 1.000 recall@3 · 30-config ablation, negative results included · live demo
|
Languages
Backend · Frontend
Databases · Infra · Tools
🧰 Expand the full arsenal
| Category | Technologies |
|---|---|
| Languages | Java Python C C++ TypeScript JavaScript (ES6+) SQL |
| Backend | Spring Boot Spring MVC Node.js Express FastAPI Flask HTMX WebSocket REST APIs |
| Frontend | React Next.js Tailwind CSS Three.js HTML5 CSS3 |
| AI / ML | PyTorch NumPy CLIP Vision-Language Models LLM Agents RAG FAISS LangChain Sentence-Transformers BM25 Cross-Encoder Reranking Gradio |
| Databases | MySQL MongoDB PostgreSQL SQLite Prisma H2 |
| Systems | Linux POSIX APIs TCP Sockets IPC Multithreading Process Management Computer Architecture Cache Simulation |
| Tools | Git GitLab Docker Postman Swagger / OpenAPI GitHub Actions |
| Project | What it is | Stack |
|---|---|---|
| 🤝 NegotiAgent | Neuro-symbolic meeting scheduler — LLM extracts formal constraints from natural-language emails; autonomous agents negotiate via Asynchronous Backtracking. 7× message reduction, 0 disagreements in 1,440 benchmark runs | Python FastAPI Groq RAG CSP/ABT |
| 🔍 CLIP Zero-Shot Anomaly | Zero-shot industrial anomaly detection & localization with CLIP — 88.5% AUROC on MVTec AD with no training on defect data. Core of my M.Tech research | Python PyTorch CLIP Computer Vision |
| 📚 LitRAG | Evaluation-first RAG over CV research papers — recall@k & faithfulness harness before the demo, 30-config ablation of chunking/embeddings/reranking. 0.957 recall@1, 1.000 recall@3. Live on HF Spaces | Python Sentence-Transformers BM25 Cross-Encoder Gradio |
| 🤖 InterviewPilot | AI-powered mock interview platform — WebSocket-driven real-time sessions, Groq API for adaptive evaluation, modular Spring Boot backend | Java Spring Boot HTMX Groq API WebSocket |
| ⚡ Distributed Task Scheduler | Master-Worker distributed system in C++ — raw TCP sockets, fork/exec, dup2, select(), dynamic load balancing. Zero external dependencies |
C++ TCP Sockets POSIX IPC |
| 🧮 Cache Hierarchy Simulator | Configurable L1/L2 CPU cache simulator in C++17 over real memory traces — LRU/FIFO/Random, write-back/through, 3-C miss classification, AMAT analysis. 32-run sweep report: direct-mapped → 4-way cuts miss rate 32.4% → 25.6%, AMAT −20.5% | C++17 Computer Architecture Python |
| 🧠 Multi-Agent AI Simulation | 8 autonomous AI agents running a simulated ISRO lunar base ("Aryabhata Station") — Stanford Generative Agents architecture with FAISS memory streams, emergent social behavior, and real-time 3D visualization | Python FastAPI FAISS Three.js Groq API |
| 🗺️ RoadmapHQ | Mission control for roadmap-based learning — track progress, goals, revisions, and consistency across structured learning paths with charts and streak tracking | TypeScript Next.js 15 Prisma SQLite Tailwind |
| 🚗 RideConnect | Campus carpool platform built for NITC students — auth, ride matching, fare splitting, responsive mobile-first UI. Deployed on Vercel & Render | React TypeScript Node.js MongoDB Tailwind |
| 🔤 Neural Net Classifier | Complete feedforward neural network from first principles — backpropagation, Xavier initialization, hyperparameter search. Zero ML frameworks used | Python NumPy Flask Matplotlib |
🔬 More Projects
| Project | Description | Tech |
|---|---|---|
| 🗓️ Cadence | Local-first planning PWA — plan tomorrow tonight, track what actually got done, score the day honestly. Live | TypeScript, Vite, Tailwind |
| 📈 Grindin | Daily check-in generator for placement prep — turns a morning form into a shareable accountability report across DSA, Spring Boot and CSE core | JavaScript |
| 🗣️ Cheppu | Duolingo-style app for spoken Telugu — hear a phrase, build it, say it, all in romanized spelling. No reading or writing required | HTML, CSS, JS |
| 📄 Paper Trail Tracker | Progress tracker for the PaperTrail project | HTML, JS |
| 💼 WAMS Backend | Enterprise workflow & approval REST API — multi-level RBAC, audit trails, hierarchical authorization | Java, Spring Boot, PostgreSQL |
| ⚔️ CodeArena | Real-time collaborative coding interview platform with WebSocket and JWT auth | Java, Spring Boot, WebSocket |
| 🔬 Multimodal Anomaly Detection | M.Tech research — multimodal anomaly detection with Vision-Language Models | Python, CLIP, PyTorch |
| ☕ JavaReflex | Drill app to build Java coding muscle memory for interviews and online assessments | JavaScript |
| 📺 FocusTube | Chrome extension that transforms YouTube into a distraction-free learning environment | JavaScript, Chrome APIs |
| 🎮 HabitQuest | Gamified Chrome extension that turns student habits into epic quests | JavaScript, Chrome APIs |
| 📚 Study Focus | Minimalist Chrome extension — blocks distractions and tracks focused study sessions | JavaScript, Chrome APIs |
| 🤖 Naukri Profile Updater | Automated job-profile freshness bot with multi-channel notifications and Docker support | Python, Docker |
| 🌐 Portfolio Website | Cosmic-themed personal portfolio with constellation animations and glassmorphism | HTML, CSS, JS |
| 💻 LeetCode Solutions | Curated algorithmic problem solutions with clean, documented code | Java |
I'm always open to discussing backend architecture, distributed systems, AI/ML research, or exciting collaboration opportunities.
If my work resonates with you, let's talk!
