Computer Engineering @ Sharif University · Vision-Transformer Interpretability
I build the retrieval and knowledge layer that turns messy, unstructured data into structured knowledge people and AI agents can actually use.
flowchart LR
A["Unstructured data"] --> B["Entity & relation extraction"]
B --> C[("Knowledge graph")]
C --> D["Embedding + graph retrieval"]
D --> E["Ranking & context building"]
E --> F["Grounded answer + attribution"]
| Role | What I worked on |
|---|---|
| Applied Scientist · food-delivery platform | Graph-based recommenders, Learning-to-Rank ranking pipelines (LambdaMART), CTR prediction, and an LLM agent that automates content-policy review |
| ML Research Intern · university ML lab | Feature attribution & mechanistic interpretability for Vision Transformers (Tuned Lens) — targeting ICLR |
| Data Scientist · ad-tech platform | Interest-based targeting & session classification to optimize ad performance |
| Project | Focus |
|---|---|
| Retrieval System for Movies | Query processing, retrieval & relevance ranking of documents — IR + deep learning |
| Map Matching & Destination Suggestion | Aligning noisy GPS traces to road segments & predicting routes — graph + sequence modeling |
| GPT from Scratch | Transformer / GPT built end to end to understand LLM internals |
| Diffusion Models | Denoising diffusion generative models, implemented and explored |
Focus
Exploring
