About
As my freshman year was getting over, my interests shifted from understanding CS fundamentals to pushing its boundaries - building things, breaking things and figuring out why. One thing led to another and before I knew it, I had gone from tutorials to research papers, production pipelines and shipped products. Since then I have researched at Cambridge and Würzburg, done multiple AI and SWE internships, built a few things worth talking about, and just started off as a Machine Learning Engineer at UnifyApps.
Work Experience
UnifyApps
Machine Learning Engineer • Gurgaon, India
Fine-tuned bge-m3 embeddings using SetFit's contrastive learning to hit 92% accuracy at under 20ms latency for an Enterprise Search LLM router. Built ToolBridge, a plug-and-play connector framework with RBAC, citations and pagination baked in, and designed a tiered core/stress evaluation philosophy for ES Agent.Predii
Applied AI Intern • Remote
Built production async LLM evaluation pipelines processing 37K+ samples end to end using MongoDB, Azure and SLURM. Benchmarked LLMs and embeddings against SME standards and authored Evaluation Framework 1.0 covering part-search, reasoning and cross-model comparison workflows.IN3TECH Computer Systems - FZE, Dubai
Software Engineer Intern • Remote
Built a Meta Glasses powered face recognition system integrated with a Flask web app and Google Search API for enriched data retrieval using LLaMA and Gemini. Developed a paperless parking solution achieving 82% license plate detection accuracy using TensorFlow, OpenCV and a MySQL backed Flask backend.
Research Experience
WüNLP, University of Würzburg
Research Intern • Würzburg, Germany
Deployed a fully automated vLLM based synthetic data pipeline on SLURM clusters producing 2M+ reasoning samples across 10+ public datasets. Built LLM-as-Judge classification workflows and optimised prompt templates across 15+ SOTA models including DeepSeek R1 and Qwen QwQ for reasoning evaluation.AFAR Lab, University of Cambridge
Research Studentship • Remote
Curated and standardised 15+ multi-party conversation datasets for robotic dialogue systems, then fine-tuned Qwen2.5-7B and Phi-4-mini using supervised fine-tuning with LoRA. Designed multi-dimensional evaluation pipelines combining human scoring and semantic analysis, with statistical testing on 1500+ samples confirming 46-60% gains in context consistency.MIDAS Lab, IIIT Delhi
Undergraduate Researcher • New Delhi, India
Worked on two projects: automated code generation using LLMs, building a 470K sample fine-tuning dataset for Qwen Coder 2.5 and evaluation pipelines with IR-level similarity analysis; and a suspect retrieval system for Delhi Police, training the FIRe model on CelebA and integrating LLaVA for multimodal identification combining images and text descriptions.
Projects
Networks, Bots, Health and Comms.
Full-stack to ML, each one scratching a different itch.
ChronoGraph
Analysed 59K+ time-stamped interactions across 1,899 nodes using NetworkX, implementing burstiness analysis, hub tracking and preferential attachment detection with an interactive Streamlit dashboard.
InvisiGate
Bot detection platform classifying behavioural signals in real time to enforce tiered responses via an embeddable JS snippet, Node.js/Express backend, React dashboard and Electron desktop app.
RecommendCare
Engineered a personalized health platform using Apple Watch data, hybrid prediction models, and RAG-based LLMs for real-time vitals monitoring and tailored health recommendations.
Drop Me a Line
Interested in connecting? Shoot me a message on LinkedIn or email me directly, and I’ll be right at ya!