AI/ML Engineer | Agentic AI · Model Evaluation · Scientific ML
I build AI workflows and research tools that people can inspect, reproduce, and extend. My work spans agent orchestration, evaluation, and learning from physical measurements, with a focus on understanding where systems fail.
When can an MRI reconstruction agree with its measurements and still be wrong?
A reproducible synthetic benchmark of reconstruction errors and the limits of independent Fourier audits. Includes a small trained CNN, open weights and data, an audit API, and an unreviewed methods preprint.
Explore the experiments · Validation catalogue · Kaggle training notebook
What do the current measurements leave ambiguous, and which measurement should come next?
A toolkit for bounded linear inverse problems that constructs conflicting admissible sources, bounds a requested quantity, and selects a next measurement by worst-case remaining ambiguity. Its certificate checker uses exact rational arithmetic and Python's standard library. The release includes a synthetic pilot and a technical note awaiting external scientific review.
Methods and proofs · Pilot results
How does a reconstruction model behave when acquisition conditions change?
A PyTorch toolkit for photoacoustic reconstruction, acquisition stress tests, and uncertainty experiments. The synthetic GPU pilot compares augmented and unaugmented training, with public checkpoints, data provenance, and reproducible evaluation. Real-measurement evaluation and uncertainty calibration are ongoing research goals.
Run the toolkit · Pilot report · Reliability checks
Selected upstream pull requests, open for review as of November 3, 2026:
| Project | Proposed improvement |
|---|---|
| AWS SAM CLI | Preserve literal backslashes during Fn::Sub substitution. |
| Cleanlab | Add outlier prediction diagnostics and update Python compatibility and CI. |
| Evidently | Fix NDCG calculation for floating-point DataFrame indices. |
| Qdrant MCP Server | Escape stored values when formatting search results. |
My experience includes AI evaluation at Handshake AI, agentic AI training at Datavalley, ML research at Dyne, and Java full-stack development through TalentNet (Wipro).
Core tools: Python, PyTorch, NumPy, pandas, scikit-learn, LangChain, CrewAI, FastAPI, SQL, Docker, GCP, and AWS. I also work with Java, Spring Boot, and TypeScript.
I'm based in India and open to remote AI/ML engineering roles, especially work on agent workflows, evaluation, and reliable ML systems. Connect on LinkedIn.