Skip to content
View MrDanial-Rafiee's full-sized avatar
  • Joined Aug 30, 2026

Block or report MrDanial-Rafiee

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
MrDanial-Rafiee/README.md

Danial Rafiee

Machine learning engineer working across computer vision, NLP and applied ML — from reconstructing damaged Akkadian tablets with an explainable ByT5 ensemble (2nd place, Innoverse Expo AI Programming Challenge) to Show-Attend-and-Tell captioning in PyTorch and retrieval-augmented question answering.

I care about work that is reproducible and honestly measured: every project below reports a baseline, states what its numbers do and do not mean, and runs top to bottom from a clean clone.

Selected projects

INNOVERSE — AI Archaeologist · 2nd place, Innoverse Expo AI Programming Challenge Explainable AI for reconstructing incomplete Akkadian texts. A ByT5 ensemble backed by retrieval from attested tablet lines, knowledge graphs and calibrated confidence — every reconstruction is labelled as an attested match, a cited parallel or a model estimate, so an inference is never presented as historical fact.

Image Captioning with Visual Attention PyTorch implementation of Show, Attend and Tell — ResNet-50 encoder, Bahdanau attention, LSTM decoder with beam search and BLEU evaluation.

Traffic Accident Risk Prediction End-to-end classification pipeline with threshold tuning and seasonal models. ROC-AUC 0.952, F1-macro 0.735.

Medical RAG Assistant Retrieval-augmented QA over a medical corpus — FAISS + sentence transformers + LangChain, with source citation and a Streamlit interface.

OCR Translator Tesseract OCR with a measurable preprocessing pipeline (projection-profile deskew, four tuning profiles), translation caching and SQLite history.

Tehran House Price Prediction XGBoost over 3,456 listings. R² 0.768 — log-transforming the skewed target beat the raw model by 4 points.

Digital Health Analyzer Screen-time and lifestyle classification with a rule-based recommendation layer and weekly trend reporting.

Also: Computer Vision Lab · Recommender Systems · TTS Studio · ML Notebooks

Tools

Python · PyTorch · Transformers · scikit-learn · XGBoost · OpenCV · LangChain · FAISS · pandas · Streamlit

Contact

Pinned Loading

  1. image-captioning-attention image-captioning-attention Public

    PyTorch implementation of Show, Attend and Tell: ResNet-50 encoder, Bahdanau attention, LSTM decoder with beam search and BLEU evaluation

    Python

  2. innoverse-ai-archaeologist innoverse-ai-archaeologist Public

    Explainable AI for reconstructing incomplete Akkadian texts: ByT5 ensemble, attested-source retrieval, calibrated confidence and evidence-based explanations. 2nd place, U2 Innoverse.

    JavaScript

  3. medical-rag-assistant medical-rag-assistant Public

    Retrieval-augmented medical question answering with FAISS, sentence transformers and LangChain, with source citation

    Python

  4. ocr-translator ocr-translator Public

    Tesseract OCR with a measurable preprocessing pipeline, projection-profile deskew, translation caching and SQLite history

    Python

  5. tehran-house-price tehran-house-price Public

    XGBoost price prediction over 3,456 Tehran listings. R-squared 0.768 with a log-transformed target

    Jupyter Notebook

  6. traffic-accident-risk traffic-accident-risk Public

    Accident risk classification with model comparison, threshold tuning and seasonal models. ROC-AUC 0.952, F1-macro 0.735

    Python