Data Scientist · ML Engineer
I'm Akhil Kanukula — a Data Scientist and full-stack builder. From NASA-funded climate forecasting with diffusion models to LLM-driven research platforms, I turn complex data into production-grade ML systems.
System Scan · Tech Stack
Experience
Cloud Leap Technologies
Jun 2026 – Present
Data Scientist
Driving data engineering and ML across multiple customer projects: analyzing CT-scan data to eliminate foreign-object scan corruption, building a UI/data platform for options trading (Baazi.ai), and engineering pipelines for AI-driven VR/XR visualization. Supporting the MD State Lighthouse Industries program on AI/ML workforce upskilling.
Platinum Business Services
Feb 2026 – Present
Data AI Engineer Intern
Design and develop use cases and MVPs for R&D AI applications in Health Care, Cybersecurity, and Quantum Computing, with a focus on Artificial Intelligence in Animal Testing Research.
Health Tech Alley
Sept 2025 – Jan 2026
Data Analyst
Built childcare demand/supply datasets, relational schemas, and Power BI dashboards to reveal shortages at the census-block level.
UMBC · NASA
Aug 2024 – Aug 2025
Graduate Research Assistant
Developed a DDPM-conditioned U-Net for monthly 2-m temperature anomaly forecasting using ERA5 GRIB/NetCDF pipelines with xarray + cfgrib.
Cyber Pack Ventures Inc.
Jan 2025 – Dec 2025
Software Developer Intern
Built an LLM behavioral-study platform with three fixed AI personalities, hallucination logic, and full-stack logging on Next.js + FastAPI + PostgreSQL.
Tata Consultancy Services
Sept 2021 – May 2024
MuleSoft Developer
Designed RAML-based REST APIs, integrated Angular front ends with SAP Commerce Cloud, and deployed services on CloudHub with CI/CD.
Robust ML Systems
My Projects
Harmonizes data from multiple satellite constellations (VIIRS, SMAP, ERA5) with Deep Learning: a U-Net for wildfire segmentation through smoke using thermal data, and an LSTM for drought forecasting via soil-moisture prediction — served through a full-stack Next.js + FastAPI app.
A U-Net-based diffusion model trained on EuroSAT Sentinel-II imagery to simulate land-cover transitions, aiding urban planning and environmental monitoring.
An NSF-funded user-study platform analyzing how people interact with hallucinating LLMs under three fixed AI personalities and task-wise flows.
DDPM-conditioned U-Net models on ERA5 reanalysis data with xarray + cfgrib pipelines, achieving improved skill over deterministic baselines on HPC.