Who am I?
Hi, Shreyash here!
I build machine learning systems, start to finish.
- Program
- B.Tech · Data Science & Artificial Intelligence
- Education
- IIIT Naya Raipur, Chattisgarh
- Current Semester
- V
- Location
- Varanasi - Uttar Pradesh - India
- Focus
- Applied Machine Learning
I'm a third year B.Tech student in Data Science and AI at IIIT Naya Raipur. The way I actually learn something is by pulling it apart down to the basics and putting it back together myself, whether that's a physics simulator, a training loop, or a retrieval pipeline. If I didn't build it, I don't really trust that I get it. Whatever I pick up along the way shows up here as TILs, and whatever I build shows up as projects. If you just want the quick version of what I'm on right now, there's a /now page for that.
School
I'm doing my B.Tech in Data Science and AI at IIIT Naya Raipur (officially Dr. SPM International Institute of Information Technology, Naya Raipur), 2024 to 2028, sitting at 8.65/10 CGPA so far. The classes I keep coming back to are reinforcement learning, machine learning, linear algebra, probability and statistics, numerical methods, algorithms, and graph theory.
Research
Lagrange-Lock is reinforcement learning for satellite station-keeping, with Dr. Avantika Singh advising. I wrote a custom Gymnasium environment on top of a CR3BP physics core, JIT-compiled with Numba so it runs around 10,000 steps a second, then trained a PPO agent for about 5 million timesteps with a reward that juggles staying in orbit, fuel use, and drift. It ended up holding Halo orbits for 5,000+ timesteps around the Earth-Moon L1 point, and there's a Three.js viewer that plays back its thrust trajectories so you can actually watch it work.
FDTD prediction via DMD is numerical modeling work, with Dr. Maifuz Ali advising. I modeled 1D FDTD electromagnetic wave propagation, then used Dynamic Mode Decomposition to predict how the field evolves, which cut the computation by roughly 10x while keeping the accuracy basically intact.
What I build with
Mostly Python, plus C++ and C when I need the speed. For the RL stuff it's PyTorch, Stable-Baselines3, and Gymnasium; for NLP and RAG it's Transformers and sentence-transformers. The full list, languages, ML stack, and all the scientific-computing and visualization tools, lives over on the /uses page.
Elsewhere
The fastest way to see what I'm up to is the projects page or the feed. If you want to reach me, email is the move.