Work

Defog AI

• Researched the effects of alignment techniques ORPO, DPO, RPO on domain‑models used for code‑generation and reasoning.
• Evaluated instruction-following capabilities of models under different reasoning and alignment traces.

NextAI

• Partnered and advised 6+ startups on technical development and deployment of personalized LLMs.
• Developed internal evaluation guidelines, datasets, and proxies to measure the performance of language models on domain‑specific tasks.

Tribe AI

• Worked on developing internal evals and benchmarks for language models.
• Orchestrated an end-to-end search tool for finding the perfect candidates to suitable open roles.

Defytrends

• Benchmarked 20+ open-source LLMs for SQL query generation, prioritizing latency, quality, and fit.
• Created a multi-GPU setup to accelerate inference and fine-tuning by 20% using DeepSpeed and vLLM.
• Developed a RAG pipeline for financial-tabular data, increasing retrieval accuracy by 40% and improving response quality by 35%.
• Successfully deployed a LLM-based financial analyst, generating investor interest from multiple international hedge funds.

IndexAI

• Spearheaded the development of a search framework to enhance enterprise search and productivity, presented at buildspace.

Dresslife

• Orchestrated a shift to AWS Kinesis, Firehose, S3, and Lambda-based microservices, enhancing data throughput by 40% and reducing latency by 20%.
• Designed and managed robust ETL pipelines, optimizing data flow for recommendation systems and increasing data processing efficiency by 30%.
• Reduced cloud computing expenses by 25% and automated ML model deployment and monitoring using AWS Sagemaker.

Locallink

• Developed a full-stack application on the MERN stack, boosting engagement by 20% in three months.
• Implemented CI/CD pipelines using GitHub, reducing deployment errors by 50% and accelerating release cycles by 35%.
• Created and maintained databases using SQL and Django, ensuring optimal latency for production workloads.

Ashoka University

• Developed an efficient radar system for object and human detection, reducing costs by 80%.
• Led 3+ publications at international conferences and journals like IEEE.

Bill and Melinda Gates Foundation

• Created large synthetic datasets (1M+ citizens) using Indian census and surveys for agent-based simulations used in disease prediction modeling.