Computer Science student building a foundation in machine learning through research, open source, and practical engineering. My interests span graph neural networks, LLMs, and mechanistic interpretability, with a long-term goal of developing production-grade AI systems grounded in strong fundamentals.
| research | mechanistic interpretability · ML security · PEFT · agentic systems |
| building | AI infrastructure · research prototypes · asynchronous backends |
| stack | Python · PyTorch · Agentic Frameworks · Vector DBs · Backend Infrastructure · Linux |
| open source | deep learning frameworks · scientific computing ecosystems |
| writing | under-the-hood-ai.hashnode.dev |
| contact | sanjana.s0528@gmail.com |
# currently
→ auditing transformer internals via sparse autoencoders
→ adversarial robustness in complex-valued GNNs [ preprint in prep ]
→ exploring the Hugging Face ecosystem