About
I'm a Data Science graduate of IIT Madras and a former AI research intern at Bud Ecosystem. I like to think about how systems model reality. Specifically, how certain systems represent and process information. Outside of that I like to play video games, listen to music and enjoy cooking.
My research interests are Mechanistic Interpretability, Deep Generative Models, and Theoretical Computer Science. I'm also drawn to digital physics, the idea that everything is fundamentally computational, and to the philosophy of mathematics...
Background
- Education
- BS, Data Science and Applications, 2022–2026, Indian Institute of Technology Madras
- Minors
- Generative AI; Economics and Finance
- Experience
- AI Research Intern, Aug 2024 – Mar 2025, Bud Ecosystem Inc.
Led R&D on the routing framework in Bud Novaria AI OS, and built Maxwell TCS v0.2, an open-source 0.4B task-complexity scorer with over 10,000 downloads.
Toolkit
- Languages
- Python, SQL, TypeScript, JavaScript, C
- Deep learning
- PyTorch, Hugging Face (Transformers, Datasets, Tokenizers, PEFT, Accelerate, Sentence Transformers), scikit-learn, XGBoost, LightGBM, CatBoost
- Post-training
- TRL (SFT, DPO, GRPO), Unsloth, distillation, quantization (GPTQ, AWQ), FSDP, DeepSpeed
- Evaluation
- DeepEval, LM Evaluation Harness, LightEval, Hugging Face Evaluate, Inspect AI, promptfoo, Ragas, Argilla
- Eval methods
- LLM-as-a-judge, golden datasets, agent evals, red-teaming, regression and fuzz testing, bootstrap confidence intervals, Cohen’s kappa
- Agents
- LangChain, LangGraph, MCP
- Retrieval
- RAG, embeddings, rerankers, hybrid search (BM25), vector DBs, pgvector, Qdrant
- Inference
- vLLM, SGLang, llama.cpp
- Interpretability
- TransformerLens, SAELens, nnsight
- Quantum
- Qiskit, OpenQASM 3.0
- MLOps
- Weights & Biases, MLflow, DVC, Airflow, Prefect, drift monitoring
- Observability
- Langfuse, LangSmith, Arize Phoenix, OpenTelemetry
- Data
- Pandas, NumPy, Polars, DuckDB, Parquet, Matplotlib, Seaborn, Plotly, PostgreSQL, Neon, Supabase, Redis
- Web
- FastAPI, Flask, Celery, SQLAlchemy, Pydantic, asyncio, React, Next.js, Electron, Vite
- Testing
- pytest, Vitest, MSW, OpenAPI contract testing
- DevOps
- Docker, Kubernetes, Linux, AWS (S3, Lambda, SageMaker), GCP, GitHub Actions, Cloudflare, Vercel, Render
- Tools
- Git, Jupyter, Google Colab, Kaggle
Recognition
- Sep 2026
- Best Capstone Project in Software Engineering, Rank #1, IIT Madras
- Apr 2026
- NPTEL Star 2026, NPTEL
- Dec 2025
- Placed 2nd of 300 on the private leaderboard, SemEval 2025 Task 11, Track A
Talks
- May 2025
- Guest speaker, World Technology Day: AI, and why high school students should not miss the chance to pursue it, All India Radio Ooty
Certifications
- Jun 2026
- SQL Intermediate, HackerRank
- Apr 2026
- Theory of Computation, NPTELQuantum Computing and Quantum Algorithms using Qiskit, NPTELProject Management, course topper (top 1%), NPTELFinancial Management for Managers, NPTEL
Involvement
- 2025 – now
- Member, Cohere Labs Open Science Community
- 2024 – 2025
- Team Leader, NebulaPioneers, Space & Astronomy Club
- 2020 – 2022
- Senior Student Secretary, Academic Honor Society, Lucknow Christian College
- 2019 – 2020
- Junior Student Secretary, Academic Honor Society, Lucknow Christian College
Writing elsewhere
-
Maxwell: TCS Lightweight SOTA for Prompt Complexity Scoring
On budecosystem.com. External link.
-
The Cost Conundrum: The Goose Chase for Cost-Effective LLMs
On budecosystem.com. External link.
Contact
Email is the best way to reach me: . I'm also on GitHub, Hugging Face, LinkedIn and ORCID.
Opinions on this site are my own.