Skip to content
Abdullah Al Rafi

Available: Open to Werkstudent & thesis roles in ML/NLP

I build NLP and document-AI systems that survive production.

ML engineer · MSc NLP at Universität Trier · previously Senior ML Engineer at bKash, Bangladesh’s largest mobile-financial service.

Production figures from bKash, 2023–2025.

01Selected work

Case studies, with the trade-offs left in.

The problem, the constraints, the decisions and the alternatives that lost, for three systems I built.

  • Fig. 01
    • Document AI
    • Triton
    • ONNX Runtime

    Bangla birth-certificate OCR

    A first-of-its-kind multilingual OCR that lifted weekly successful onboarding up to 3×.

    Read case study

  • Fig. 02
    • RAG
    • FastAPI
    • Hugging Face

    Document RAG system

    Ask questions of any PDF and get streamed, document-grounded answers. Uploads never touch disk.

    Read case study

  • Fig. 03
    • Automation
    • PostgreSQL
    • Telegram

    Lazy Termin

    Watches a Termin booking page and alerts subscribers on Telegram. An outbox means no alert is lost or sent twice.

    Read case study

02Experience

Where the numbers come from.

  1. –

    Senior ML Engineer, Data Science · bKash Limited (opens in a new tab)

    Raised birth-certificate OCR performance by 8% with new document types, lifting weekly successful onboarding up to 3×.

  2. –

    ML Engineer, Data Science · bKash Limited (opens in a new tab)

    Built a first-of-its-kind multilingual birth-certificate OCR and served it at 100k+ requests per week.

  3. –

    Junior Data Scientist, Data Science · Aequisym AI (opens in a new tab)

    Fine-tuned Bengali ASR with NVIDIA NeMo and built a gold-standard medical-domain dataset.

03Contact

Hiring, supervising a thesis, or building something with ML? Let’s talk.

Based in Trier, Germany (CET) · replies within 48h.

Ways to reach me