Experience
Independent Researcher
2025 — PresentMachine Learning & NLP
Investigating when the spectral and geometric character of learned representations determines what a system can and cannot do — and what that implies about when pipelines will fail.
- Designed AnchorSum, a modular training-free summarization pipeline combining entity-guided anchor extraction and dual-mode NLI faithfulness auditing.
- Empirically documented verifier exploitation in prompt-only iterative refinement and proposed a multi-metric triangulation protocol for its detection.
- Outperformed fine-tuned encoder-decoder baselines (BART, PEGASUS, PRIMERA) across all LLM-as-judge dimensions using a zero-shot inference-time pipeline.
- Introduced the spectral saturation index S(K) — a closed-form, label-free stopping rule for few-shot label acquisition, validated across 49 real tasks (binary, 5-way, 10-way) and three frozen backbones.
Suvidha Foundation
Sep 2025 — Nov 2025Machine Learning Intern
- Researched multi-document abstractive summarization with transformer-based models under mentorship of a PhD in NLP.
- Built a proof-of-concept summarization system incorporating hierarchical context aggregation across multiple source documents.
- Participated in community outreach programs supporting underserved populations alongside technical work.
Harvard Computer Society AI Bootcamp
2025Artificial Intelligence Fellow
- Completed an intensive AI fellowship collaborating with Harvard students, focused on frontier AI systems, modern architectures, and real-world deployment challenges.
- Implemented a Transformer from scratch: sinusoidal positional encoding, multi-head self-attention, cross-attention, and position-wise feed-forward networks.
- Explored Reinforcement Learning foundations: MDPs, Value Iteration, Policy Iteration, SARSA, Q-Learning, and Multi-Armed Bandits.
- Completed hands-on projects: fine-tuned BERT, built a mini-LLM, implemented Vision Transformers, and constructed neural networks from first principles.