AI/ML contributor & benchmark author · Kaggle Notebook Expert · CSE (AI & ML)
I build ML systems, compete in machine learning challenges, and design new problems for other people to solve. Through Project Eris at Shipd by Datacurve, I work on both sides of the leaderboard: authoring benchmarks and developing solutions.
My workflow: understand the data → build a reliable validation setup → experiment → ship. I'm especially interested in NLP, computer vision, structured prediction, and the engineering that makes models useful.
|
01 / RACE SIMULATION
A Formula 1 simulator combining six ML models, historical telemetry, and Monte Carlo race scenarios, streamed lap by lap to a React dashboard over WebSockets.
|
02 / OBSERVABILITY
A Kubernetes observability dashboard connecting CPU anomaly detection, eBPF dependency graphs, and natural-language questions through a local LLM.
|
|
03 / MOLECULAR ML
Molecular toxicity prediction pairing a graph attention network with an 8-qubit quantum circuit. Evaluated across 12 Tox21 assays using scaffold splits, with a classical baseline and atom-level explanations.
|
04 / KNOWLEDGE GRAPHS
Turning academic PDFs into knowledge graphs and assessment questions, with graph-selected distractors, deterministic Cypher checks, and a dashboard for evaluating generated questions.
|
Shipd by Datacurve · Project Eris · AI/ML contributor & benchmark author · March 2026–present
- End-to-end authoring: datasets, held-out splits, graders, and submission validation. Authored 35+ approved ML benchmark challenges, with new challenges added regularly.
- Evaluation design: leakage controls, compositional splits, metrics that account for rare classes, and baseline comparisons that test for shortcuts.
- Reproducible pipelines: preprocessing, model logic, and evaluation built against explicit correctness and edge-case rubrics.
- Shipd · Project Eris: Top 15 and climbing on the all-time contributor leaderboard.
- Kaggle Competition Bronze Medalist · March Machine Learning Mania 2026: 232 / 3,462 teams (top 6.7%). Also a Notebook Expert, sharing reproducible ML experiments and analysis on Kaggle.
- Competitive ML: experiments across language, vision, and structured data, with validation and error analysis driving the next iteration.
- Challenge design: new task formulations on niche datasets, with careful evaluation and reproducible pipelines.
- Open source: building a contribution practice around the tools I use, starting with focused fixes and useful tests.
Models & experiments PyTorch PyTorch Geometric timm scikit-learn LightGBM Optuna
LLMs & graphs Transformers GraphRAG LangChain LlamaIndex Neo4j
Systems & data Python SQL FastAPI React Streamlit Docker Kubernetes
Education: B.Tech in Computer Science & Engineering (AI & ML), Bengal Institute of Technology, Kolkata (MAKAUT). Expected graduation: May 2027.
A little beyond the code
Based in Kolkata. Studying computer science with a focus on AI and ML. Usually alternating between model experiments, photography, anime, and a run or a lifting session.
Curious problems. Careful experiments. Useful software.