Skip to content
View Bekafi01's full-sized avatar

Highlights

  • Pro

Block or report Bekafi01

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. 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
Bekafi01/README.md

Hi, I'm Bereket Fikre πŸ‘‹

AI / ML Engineer

Building production LLM systems, custom Transformer architectures, and reliable applied ML.

LinkedIn Hugging Face Email Open to Work


🧭 About Me

I am an AI / ML Engineer specializing in Natural Language Processing, Large Language Models (LLMs & RAG), and predictive statistical systems. My experience ranges from engineering from-scratch PyTorch Transformers and contributing to foundation models (e.g. AmhariXLM-0.5B at Fideliz.ai) to building production RAG systems with rigorous evaluation benchmarks and sub-millisecond vector retrieval.

  • 🧠 Foundation Models & NLP β€” built from-scratch 60.5M PyTorch Transformers, contributed to AmhariXLM-0.5B at Fideliz.ai, tokenization pipelines (BPE/SentencePiece), and Ethiopic homophone folding
  • πŸ” RAG & GenAI Systems β€” deterministic neighbor expansion for multi-page textbook retrieval, Pydantic structured citations (CitedAnswer), and RAG Triad automated evaluation harnesses
  • πŸ“ Forecasting & Statistical ML β€” hierarchical time series (MinT reconciliation), uncertainty estimation (Split Conformal Prediction), and drift monitoring (PSI)
  • πŸš€ Delivery & MLOps β€” containerization with Docker, API serving with FastAPI/Flask, interactive UIs with Streamlit, and Hugging Face model deployment

πŸ“Œ Featured Engineering & Research Projects

A production-grade conversational RAG platform indexing 2,400+ national textbook chunks. Features deterministic neighbor expansion for multi-page retrieval, Pydantic citation guardrails, and an automated LLM-as-a-Judge benchmark (100% recall, 91.7% correctness).

Live Demo Gemini ChromaDB LangChain

A from-scratch PyTorch bidirectional Neural Machine Translation system (60.5M params) trained on 4.4M+ sentence pairs with joint 32k BPE tokenization, Ethiopic text normalization, and beam search decoding.

Hugging Face Model PyTorch Transformer FLORES-200

Production forecasting platform featuring hierarchical MinT reconciliation, distribution-free uncertainty intervals via Split Conformal Prediction, Tweedie LightGBM for zero-inflated demand, and PSI drift telemetry.

LightGBM MinT Conformal Prediction PSI Drift

End-to-end recommendation engine combining collaborative filtering and semantic embeddings with sub-millisecond FAISS vector similarity retrieval, FastAPI serving, and an explainable Streamlit UI.

FAISS FastAPI Streamlit Docker


πŸ› οΈ Tech Stack & Capabilities

🧠 Machine Learning, LLMs & NLP

πŸ’» Languages & Core Data

πŸš€ Serving & Delivery

πŸ“Š Evaluation, UI & Explanability


πŸ“Š GitHub Activity

GitHub stats Top languages

Open to AI/ML Engineering & Applied Research opportunities and collaborations.

πŸ“« Reach me at bereketfi01@gmail.com or connect on LinkedIn

Pinned Loading

  1. movie-recommender movie-recommender Public

    Hybrid movie recommendation platform with semantic embeddings, collaborative filtering, explainability badges, and an interactive Streamlit UI.

    Python

  2. internship-readiness-platform internship-readiness-platform Public

    AI-powered full-stack platform that evaluates student readiness for technical internships by parsing resumes, benchmarking profiles against real market demand, identifying skill gaps, and generatin…

    Python

  3. bank-retention-decision-engine bank-retention-decision-engine Public

    End-to-end banking churn prediction, probability calibration, and deposit-at-risk financial optimization engine with FastAPI and Streamlit.

    Jupyter Notebook 1

  4. english-amharic-transformer-nmt english-amharic-transformer-nmt Public

    From-scratch Transformer NMT for English ↔ Amharic (PyTorch). Data pipeline β†’ BPE β†’ training β†’ beam search β†’ FLORES-200 eval β†’ live demo.

    Python

  5. freshman-modules-rag freshman-modules-rag Public

    A production-grade conversational RAG tutor for Ethiopian university freshman curricula, featuring zero-latency neighbor-expansion retrieval and exact page citations.

    Python