I'm a final-year CS (AI/ML) student who ships full products, not notebooks — RAG pipelines, safety-critical LLM systems, multi-agent research tools, and the occasional Gen Z aura calculator. If it involves an LLM doing something useful under the hood, I've probably built it.

I'm currently looking for Software Engineering, AI Engineering, and Product roles where I can turn "the model said so" into something explainable, shippable, and actually used.

Currently:

• 🚀 Product Management & GTM Intern @ Morph Systems
• 🧠 Selected for Amazon ML Summer School 2026

• Languages: Python, JavaScript, Java, C++, R
• LLMs & Agentic AI: Ollama (Llama 3), Google Gemini, Groq, Prompt Engineering, RAG, Multi-Agent Systems, LangChain, Google Genkit, Amazon Bedrock & Lex, RLHF alignment
• RAG & Vector Search: FAISS, BM25 (hybrid retrieval), sentence-transformers
• ML / Deep Learning: PyTorch, scikit-learn, XGBoost, LSTM + Attention, spaCy, FLAIR, DeBERTa, Gensim
• Explainable AI: SHAP, counterfactual XAI, bias/fairness auditing
• Voice & Speech: AssemblyAI
• Web Development: React, Node.js, Express, FastAPI, Vite, Tailwind CSS, Framer Motion, Leaflet.js
• Databases: MongoDB, Firebase Firestore, DynamoDB, FAISS, SQLite, Redis, PostgreSQL (Supabase)
• Cloud & DevOps: AWS (Lambda, API Gateway, Cognito, DynamoDB, Bedrock, S3, CloudWatch), Firebase, Docker, Hugging Face Spaces, Render, Vercel
• Testing & Tools: Git, Selenium, pytest, Postman

• AirMind AI: urban air intelligence operating system — transforms cities from reactive monitoring to proactive intervention with XGBoost forecasting, pollution attribution, intervention planning, and dual AI copilots for citizens and commissioners. Graceful degradation ensures it never breaks during demos.
• PulseOps: AI-assisted cluster observability platform — complete detect → alert → incident → diagnose loop with 32 REST endpoints, WebSocket real-time updates, Prometheus metrics, and RAG-grounded AI diagnosis. Zero-credential fallbacks for immediate demo capability.
• NEXUS — Consultancy AI: a local-first AI market-intelligence platform — 8 research agents + 4 enterprise modules (market research, M&A due diligence, live risk monitoring) fusing classical NLP, deep learning, and a local LLM. Zero cloud dependency, zero data leaving the machine.
• NurtureAI: a bilingual AI parenting assistant where safety doesn't trust the LLM — a deterministic rule engine can bypass or override the model entirely on emergency queries. RAG-grounded advice, voice input, English + Arabic.
• ComplianceAI: call-center compliance monitoring that actually understands Tanglish and Hinglish. Every AI decision comes with a receipt — a direct transcript quote, not just a score.
• Product Intelligence Engine (Mercado): 4 microservices, 4 parallel Gemini agents, and a financial risk model that turns Reddit complaints into a boardroom-ready PDF — all on free-tier infra because I refuse to pay for compute I don't need yet.
• Consumer Intelligence Platform: segmentation, churn/LTV prediction, SHAP explainability, and bias auditing in one pipeline — with an LLM writing the customer personas so nobody has to stare at cluster IDs.
• Network QoE Prediction System: predicts telecom churn before the customer even complains. LSTM + XGBoost dual-model, SHAP tells you exactly which KPI is to blame.
• FocusWin: a to-do app that got out of hand — 10-level priority scale, honest analytics (no vanity heatmaps), and a real-time team Kanban spin-off called Spaces.
• OneStop 25: year-end reflection app for 2025→2026 — aura calculator, IN/OUT lists, AI-personalized manifestations. Brainrot-coded on purpose, so I can actually track my 2026 goals without opening a spreadsheet.
• SafeSpace: AI journaling with an actual feedback loop — Dr. Luna (Groq/Llama 3) reads every entry and returns sentiment, mood, and a nudge. Streaks, heatmaps, gamification, the whole loop.
• CivicBot: report a pothole over WhatsApp, get it triaged by an LLM before a human even looks at it. 100% serverless AWS, because I don't do idle servers.

• FibFormer — fusing hand-crafted Fibonacci trading heuristics into a differentiable Temporal Fusion Transformer for BankNifty/Nifty forecasting, validated with the Deflated Sharpe Ratio. (paper under IEEE review)
• Adv-Audit — neuro-symbolic explainable AI for real-time ad bidding: SHAP attribution, bias detection, and human-in-the-loop rule correction.
• HELM-Rank — an exam-grading pipeline combining OCR, knowledge-grounded RAG, Elo ranking, and RLHF so teachers can actually align the grader instead of trusting it blindly.
• GEO: Algorithmic Moat — quantifying whether LLMs systematically favor incumbent brands over startups, using SHAP and counterfactual XAI to prove it mathematically.

• AI Agents & Automation: if I do a task twice, I automate it — efficiency is just laziness with planning.
• GenAI & XAI: I build LLM-powered tools that explain themselves, because "the model said so" is not an explanation.
• Safety-critical AI: the interesting part of an AI system isn't the happy path, it's what happens when the model is wrong.
• Full-stack product thinking: an AI pipeline nobody can use is just a very expensive script.
• Cloud & Serverless: strong believer in other people's computers — Lambdas should live fast and die cheap.

• B.Tech in Computer Science, AI/ML Specialization | VIT Chennai (2023–2027) | CGPA: 9.25
• Selected for Amazon ML Summer School 2026
• Machine Learning Specialization — DeepLearning.AI & Stanford University (Coursera)
• MERN Full Stack Certification — Ethnus & CDC, VIT University | GeeksForGeeks
• Ericsson Edge Academia '25
• McKinsey Forward Learners Program
• Relevant Coursework: DSA, OOPs, OS, CN, DAA, Software Engineering, DBMS, Cloud Architecture, Machine Learning, Deep Learning, Explainable AI, NLP, Compiler Design

My Github Skill Sync

"you can always do stuff"