Tracks every new arXiv paper in computer science each day, picks the weekly top papers with a transparent, evidence-backed score (citations, community upvotes, code adoption, venue) and turns any topic into a briefing: what’s new, emerging terms, research directions and growth.

  • Python
  • FastAPI
  • SQLite FTS5
  • Sentence embeddings
  • arXiv API

A weekly, fully automated statistical and ML analysis of how people talk about AI and robotics on Bluesky, Reddit, Hacker News and Mastodon: negative-binomial GLMs, burst detection, embedding-based topic clustering and sentiment, published as a static dashboard.

  • Python
  • Statistics
  • NLP
  • Topic modelling
  • GitHub Actions

A local computer-vision system that relates facial expression (YuNet + MobileFaceNet on OpenCV DNN, ~26 ms per frame on CPU) to what is in focus on the desktop. Dashboard, daily/weekly reports and an API made for an LLM “mentor”. No frame is ever stored.

  • Computer vision
  • OpenCV
  • FastAPI
  • SQLite
  • Docker

My B.Sc. final project: find Digikala products by describing them in Persian or uploading a photo (CLIPfa embeddings in Weaviate), then ask a local Qwen2.5 shopping assistant running on llama.cpp, CPU only. Later rebuilt as a containerised, tested application.

  • CLIP
  • Vector search
  • Weaviate
  • llama.cpp
  • Streamlit

More projects

paperx

2024

A tool for reading and querying research papers: GROBID parses the PDF into structure and a Llama 2 model answers questions about it.

Mining association rules from a dataset and then hiding the sensitive ones with sanitisation techniques. Data-mining course project.

A Pygame visualisation of the N-Queens problem solved by min-conflicts local search, built so other solvers (backtracking, genetic algorithms, simulated annealing) can be plugged in.

Implementations from my data-structures course: linked lists, stacks, queues, trees, sparse matrices, polynomial arithmetic and expression conversion.