Feb 2025 – Jun 2025 · 4 mos · Amherst, Massachusetts, United States

• Academic open-source research collaboration through UMass industry mentorship program.
• Latent Traits and Cross-Task Transfer: Deconstructing Dataset Interactions in LLM Fine-tuning – Accepted to *SEM 2025 (co-hosted with EMNLP) (https://arxiv.org/pdf/2509.13624)
• Collaborating with Anthropic on understanding and evaluating cross-domain generalization offine-tuned LLMs, utilizing PEFT methods (LoRA, QLoRA) to analyze novel interactions betweentask-specific datasets and out of domain industry benchmarks. Designed experiments and fine-tuned over500 adapters on candidate datasets using LlamaFactory, Unsloth, HuggingFace, TRL for training, andvllm for inference. Ran experiments on the Unity HPC Cluster and Llama Model Family (1B - 70B).

Feb 2025 – May 2025 · 3 mos

• TA for COMPSCI 627 Fixing Social Media

Sep 2021 – Jun 2023 · 1 yr 9 mos · Pune, Maharashtra, India · Hybrid

• Contributed to the research and development of AWS CodeGuru, a large-scale code analysis engine.
• Built scalable, fault-tolerant services integrated with AWS infrastructure, using Lambda and CloudWatch for event-driven processing and real-time monitoring.
• Owned a Python code analysis utility (1M+ LoC) upgrade (200 versions), reducing analysis time by 40%.
• Increased engine code coverage by 12% by developing and documenting APIs for efficient traversal of Python and TypeScript code graphs, integrated with CodeBuild and CodePipeline for CI/CD.
• Reduced timeouts by 17% by redesigning sharding strategies to minimize latency in processinglarge-scale code repositories, while maintaining balanced code coverage and throughput efficiency.
• Identified key performance improvements by developing a benchmarking pipeline using SQL, Pandas, andDocker to compare against competitors on millions of records, with Matplotlib for visualization.
• Designed and implemented modular components with a Test-Driven Development (TDD) approach,backed by comprehensive unit and regression tests based on production scenario

May 2020 – May 2021 · 1 yr · Pune, Maharashtra, India

• Led end-to-end development of the Alphaponics Controller, an automated system for hydroponic farms.
• Increased crop yield by 23% by developing a predictive ML-system using Random Forest and XGBoostbased on temporal and environmental data to optimize nutrient dispensation.
• Engineered sensor-actuator systems using C++ for precise control of pH, EC, nutrient supply, andoxygen saturation, reducing human intervention by 84% and improving resource utilization by 11.5%.
• Reduced critical failures by 56% by implementing a real-time autonomous system to proactively identifyanomalies in vital parameters, trigger automated actuator corrections, and issue immediate user alerts.
• Created dashboards for highlighting key metrics and enabling faster decision making using Grafana.
• Built strong client relationships by ensuring high deployment standards through system installation,sensor calibration, and rigorous onsite testing.

Mar 2020 – May 2020 · 2 mos · Pune, Maharashtra, India

• Led development of driver drowsiness detection system for ride-hailing companies, resulting in an invitation at the IEEE PuneCon 2020 and publication in IEEE Xplore.
• Designed ML model for eye-closure detection using Fast R-CNN, Keras, Numpy, Pandas and developed solutions for detecting yawning and head tilt using face mapping.

Jun 2019 – Aug 2019 · 2 mos · Pune

• Prototyped ML applications to determine application feasibility.