Open Source ML Contributor at Google Summer of Code (GSoC) 2026 (2026-06 – Present)
- Develop and optimize semantic segmentation architectures (U-Net, SegNet) in PyTorch on multi-temporal satellite imagery, increasing coastline reconstruction accuracy by 16% across 1000+ km of Alaskan territory.
- Build longitudinal time-series forecasting models on large-scale remote sensing data arrays, automating environmental tracking pipelines and reducing regional erosion predictive errors by 11%.
Machine Learning Engineer Intern at Wadhwani AI (2026-01 – 2026-06)
- Data Pipeline Infrastructure (AWS/GCP): Architected an enterprise agricultural knowledge base utilizing the Mistral OCR API, orchestrating automated cloud-native ETL jobs via AWS S3 and GCP Cloud Functions to ingest 8,000+ complex multi-modal documents.
- Production Deployment (FastAPI/vLLM): Containerized and deployed a production-grade agentic RAG pipeline using FastAPI, vLLM, and Docker to serve live agronomic advisory queries, achieving an end-to-end pipeline latency of ∼5 seconds.
- MLOps & Experiment Tracking (WandB): Systematically benchmarked 5+ distinct RAG configurations and open-source LLM backends across 1,500+ golden evaluation queries; monitored retrieval precision and generation alignment end-to-end via Weights & Biases (WandB).
- Algorithmic Optimization: Developed a hierarchical chunking strategy and conducted retrieval model fine-tuning, which boosted target context retrieval accuracy by 18% and final LLM generation correctness by 8%.
Undergraduate Researcher at Indraprastha Institute of Information Technology, Delhi (2025-10 – 2025-12)
Advisor: Dr. Pankaj Jalote
- Implemented bidirectional program slicing on C/C++ codebases, reducing vulnerability search space by 42% across 1.5M+ lines of production code.
- Fine-tuned Large Language Models (LLMs) on sliced code representations to classify vulnerabilities, achieving a 25% improvement in F1-score detection performance over established baselines.
- First-authored a research paper detailing the LLM-driven vulnerability detection pipeline, evaluating models on 5,000+ code vulnerabilities, currently under peer review at an A*-ranked conference.
Research Intern at AI Institute, University of South Carolina (AIISC) (2025-01 – 2025-10)
Advisor: Dr. Amit Sheth
- Engineered an agentic multi-modal AI execution pipeline to parse complex lecture slides, autonomously generating difficulty-graded quiz questions and assessments with a 94% factual accuracy rate against the source material.
- Programmed a personalized learning engine driven by hybrid reinforcement learning for real-time difficulty adaptation, integrating Llama-3.2 summary pipelines to boost active student retention by 19%.
- Co-authored system-architecture paper "PAL: Personal Adaptive Learner", peer-reviewed and accepted for publication at AAAI 2026 (Certificate).
Teaching Assistant - Reinforcement Learning at Indraprastha Institute of Information Technology, Delhi (2025-08 – 2025-12)
- Instructed 80+ undergraduate students in deep reinforcement learning theory (PPO, GRPO, Q-learning) and guided development of PyTorch-based agent training frameworks, yielding a 92% positive instructional rating.
Research Intern at IIT Patna (2025-05 – 2025-10)
Advisor: Dr. Sriparna Saha
- Pioneered a customized GRPO (Group Relative Policy Optimization) alignment architecture, optimizing a binary reward function (0/1) for exact ground-truth matching, which secured a 27% exact-match accuracy boost on a complex cultural QA dataset (Certificate).
Undergraduate Researcher at FLaME.nlp Research Lab, IIIT Delhi (2024-01 – 2025-06)
Advisor: Dr. Md. Shad Akhtar
- Designed a novel architecture based on knowledge distillation and Chain-of-Thought (CoT) reasoning for conversational AI models, successfully improving classification accuracy by 19% on the F1-score.
- Curated a specialized dataset of 15,000 annotated dialogue records, constructing an automated quality-control validation pipeline using Pandas that programmatically cut label noise by 14%.
- Co-authored "Measuring What Matters: Assessing Therapeutic Principles in Mental-Health Conversations", accepted for an Oral Presentation at the ACL 2026 Main Conference (Certificate).
