Jeet Shah
Software engineer & Columbia University MSCS student
New York City
I am a computer science graduate student working across production software, data systems, and applied machine-learning research. I previously worked part-time with Elevate Fitness and Voltek AI, through August 2026. I'm currently implementing Nilakan with Prajas Naik, based on language fundamentals designed by Prof. Aamod Sane. This website is an archive of my projects, research, and writing—and a place to experiment with design.
experience
Columbia Software Solutions
Oct 2026 - Present- Student-run organization building free, custom software for nonprofits and small businesses in New York City.
Columbia University
Fall 2026- Hold office hours, grade coursework, facilitate discussions, and support assignments for COMS2702 · AI in Context, a course with approximately 180 enrolled students.
Elevate Fitness
Dec 2025 - Aug 2026- Built and maintained Movynn, a React Native and Expo fitness application backed by Convex and published on the Apple App Store and the Google Play Store in India.
- Implemented native Apple Sign-In and secure session recovery, including nonce verification and first-login profile persistence.
- Built multi-ticket booking and Razorpay payment recovery and reconciliation, requiring server-confirmed capture before order fulfillment.
- Added account-deletion and pseudonymization flows, automated app and backend tests, continuous-integration checks, and release documentation.
Voltek AI
Oct 2024 - Aug 2026- Led frontend development for an internal battery-research platform used by 10 - 100 Nanoloy users.
- Redesigned PostgreSQL process storage from process-specific, join-heavy tables to three indexed canonical tables with transactional writes and backfill validation, reducing observed material-query latency from 3-20 seconds to under 500 ms.
- Built the Next.js orchestration layer for a battery-research assistant supporting OpenAI, Anthropic, Gemini, and xAI, with internal-document retrieval, web search, citations, streamed responses, and chat history.
- Co-maintained NDAX and Neware ingestion pipelines, including file reconciliation and defensive parsing for incomplete file sets.
Centre for Interdisciplinary AI, FLAME University
Jan 2024 - May 2025- Configured and administered a multi-user GPU compute server for research and model training, including user management, drivers, storage, and environment isolation.
- Co-developed ConvGRU, conditional-GAN, and ConvLSTM training pipelines and satellite-data ingestion tooling for precipitation nowcasting.
- Supported team kaubega's Weather4Cast 2024 and 2025 submissions through model development, data-ingestion tooling, and GPU infrastructure.
Space Applications Centre, ISRO
May 2023 - Aug 2023- Developed ConvLSTM precipitation-nowcasting models using INSAT-3D satellite data in an air-gapped HPC environment.
research
arXiv preprint. Co-author. Team kaubega placed second in the Weather4Cast 2025 cumulative-rainfall task.
arXiv preprint. Co-author. Team kaubega placed first on the Weather4Cast 2024 core-challenge leaderboard.
Precipitation Nowcasting Using ConvLSTM with INSAT-3D Satellite Data over the Indian Subcontinent
2025FLAME Scholar's Program undergraduate thesis supervised by Kaushik Gopalan. The ConvLSTM forecast reduced six-hour RMSE by 30.7% compared with a Lucas-Kanade optical-flow baseline.