Applied AI / ML Engineer
|
Physics-informed ML ยท Graph learning ยท Agentic systems ยท Production-minded validation
Turning domain constraints into testable ML systems, from physics-grounded condition monitoring to graph learning and agentic workflows.
Designing causal validation, leakage-safe benchmarks, interpretable baselines, and promotion gates before models reach an interface.
Architecting Flask and FastAPI services, reproducible data pipelines, REST APIs, and regression-tested analytical workflows.
Building responsive interfaces that explain model evidence, limitations, and decisions instead of presenting opaque scores.
A GraphSAGE permeability predictor for pore networks extracted from micro-CT data. The research compares learned topology against Kozeny-Carman physics and uses pore heterogeneity to choose the appropriate regime instead of claiming one model always wins.
A private, three-plant condition-monitoring program for centrifugal compressors. I built physics-grounded fouling indicators, plant-specific instability review queues, causal maintenance-event logic, and operator-facing diagnostics while keeping unsupported forecasting and surge claims behind explicit scientific gates.
An agentic threat-intelligence prototype that combines rolling market anomaly signals with structured Gemini workflows. I built the backend and orchestration path, including schema-constrained outputs, proxy routing, API-key isolation, and model fallback behavior.
A team-built DePIN microgrid prototype combining a simulated 15-node energy network, continuous-control DDPG agent, WebSocket command center, EVM settlement, and Groth16 proof flow. My contribution focused on the Python AI engine, backend integration, and 3D visualization.
An internship prototype for institutional kitchen operations, connecting menu planning, bill-of-materials calculations, inventory, and meal-attendance forecasting in one FastAPI workflow. The case study focuses on the system and decision flow rather than claiming unverified waste savings.
A mobile-first React and FastAPI prototype for exploring crop-yield scenarios across Odisha districts. The current public build focuses on accessible inputs, confidence-range presentation, and service architecture; model validation remains future work.
Built and validated a private condition-monitoring platform for centrifugal compressors across three plants. The work spans data science, thermodynamic feature engineering, ML evaluation, causal time-series analysis, Docker-based service workflows, operator diagnostics, and regression-tested Flask APIs.
Built a food-optimization and cost-management program for a Sodexo canteen, connecting menu generation, bill-of-materials calculations, inventory and cost workflows, and meal-attendance forecasting through FastAPI and SQL.
Karjat, Maharashtra
AI & ML UndergraduateApplying to the MLH Fellowship and open to collaborative software engineering, applied ML, and open-source work.