Architected a full-stack AI platform automating job applications across 8 portals.

Building the core engineering backbone of an AI-powered job automation platform. Architected a multi-repository system spanning a Next.js frontend, Vertex AI-backed intelligence services, and Puppeteer automation infrastructure with encrypted credential handling, Redis/Bull job queues, and live application tracking.

8 job portals automated

3-repository system

AES-256 credential vault

Promoted AI-powered search and responsible AI use across campus.

Selected as an official Campus Ambassador for Perplexity AI. Led sessions on AI-powered search, responsible AI usage, and practical LLM workflows for research and productivity.

Campus-wide sessions

Responsible AI advocacy

Hands-on product adoption

Grew the community from 12 to 80+ active members.

Led technical programming for the university Linux community, delivered 15+ workshops, mentored 50+ students in open source contribution, and created repeatable learning pathways that outlived the role.

Community growth: 567%

15+ workshops delivered

50+ students mentored

Reduced manual address processing effort by 87% with a production ML pipeline.

Developed a production-grade multilingual address parsing system for one of India's largest power utilities. Built and deployed an IndicBERTv2-CRF pipeline, handled annotation strategy, evaluation, ONNX optimization, and Cloud Run deployment.

94.3% F1 accuracy

87% efficiency improvement

Production deployment for 3.5M customers

Production NER system for parsing unstructured Indian addresses into structured components using IndicBERTv2-CRF. 94.3% F1 score with <25ms inference.

During my ML internship at BSES Delhi, I built a Named Entity Recognition system for parsing unstructured Indian addresses into 15 structured fields — house number, floor, block, gali, colony, area, sector, pincode, city, and more. The model uses IndicBERTv2 (pretrained on 20B+ tokens of Indian language text) with a CRF layer for sequence labeling, trained on hand-annotated Delhi addresses in both Hindi and English. It hits 94.3% F1 on a balanced test set with sub-25ms inference per address. I added a rule-based post-processor with a Delhi locality gazetteer for corrections, optimized with ONNX runtime, and deployed via FastAPI on Google Cloud Run with auto-deploy via Cloud Build.

End-to-end digital platform with React frontend and Google Sheets integration for live schedule management. Served 3000+ participants.

As Digital Systems Lead for BITS Tech Fest, I led the end-to-end design and operations of a multi-club tech festival platform serving 3000+ participants across 50+ events. The project involved a React frontend integrated with Google Sheets for direct content management and rapid deployments. The highlight was engineering a live schedule system that allowed direct updates in Sheets, reducing schedule-change turnaround from hours to minutes. Implemented automated testing pipeline and monitoring systems, achieving 99.9% uptime during peak traffic.

Model Context Protocol server that transforms AI-generated text into authentic, human-like content — eliminating AI slop patterns.

TrueVoice is a Model Context Protocol (MCP) server that helps AI systems generate genuinely human-sounding text. It analyzes content for common AI slop indicators — filler phrases, generic structures, robotic tone — and provides actionable rules and examples for producing authentic writing. Built with TypeScript and the MCP SDK, it integrates directly into AI workflows as a tool server, offering real-time text analysis, human writing guidelines, and slop pattern detection. The server ships with categorized examples across phrases, structure, and tone to help AI systems learn what to avoid.