Skip to content
View MehdiBedja's full-sized avatar
2 followers · 3 following

Block or report MehdiBedja

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
MehdiBedja/README.md

Mahdi Bedja

AI engineer based in Paris. I graduated from ESI Algiers (engineering degree in computer science) and I'm currently doing a Master 2 in Intelligent Systems at Sorbonne Paris Nord. I'm looking for a final-year internship (PFE) in applied AI starting March 2027.

Portfolio: mehdibedja.github.io · LinkedIn · mahdibedja@gmail.com

Work

AI Engineer, BM TECH (Algiers, Nov 2025 – Aug 2026)

BM TECH makes software for medical laboratories and hospitals. The code below belongs to the company, so it isn't on this profile, but this is what I worked on:

  • Detection, segmentation and classification of 14 lung pathologies on chest X-rays (YOLO, CNNs), and a separate bone fracture detection pipeline.
  • French medical speech recognition: I built a synthetic dataset by cloning physicians' voices with Qwen3 from real radiology reports, then fine-tuned Whisper. 3% WER on a held-out synthetic test set.
  • Real-time medical dictation on top of WhisperLiveKit / faster-whisper, with fuzzy matching and voice commands. 8% WER in live streaming.
  • An agentic RAG system (LangChain, locally hosted Qwen models) for answering questions over patient records and clinical documents.
  • Semantic matching of lab test results across hospitals using dense embeddings of the LOINC dataset.
  • Packaging the services with Docker, with structured logging and monitoring.

Engineering intern, CM Consulting (Algiers, summer 2024): a Python tool that extracts and structures data from PDF tenders (spaCy, pdfplumber) and generates formatted reports.

Projects

Mobile pest detection (engineering thesis). I benchmarked YOLOv7 to v11, SSD and Faster R-CNN on a curated 13-class subset of IP102, picked YOLOv11s (93.8% mAP@50), and shipped it as a TFLite model inside an Android app that works offline. The app syncs with a Django backend.

Smaller projects:

Skills

Python, PyTorch, TensorFlow, Hugging Face, LangChain, Whisper, YOLO, TFLite, Django, PostgreSQL, Kotlin, Docker.

English (C1, IELTS 7.5), French (C1, TCF), Arabic (native).

Pinned Loading

  1. pest-detection-android pest-detection-android Public

    Offline pest detection Android app: on-device YOLOv11s TFLite, Jetpack Compose, Room, EN/FR/AR

    Kotlin 1

  2. pest-detection-api pest-detection-api Public

    Django REST API for the pest detection app: token & Google auth, offline-first detection sync, Cloudinary, PostgreSQL

    HTML 1

  3. pest-detection-computer-vision pest-detection-computer-vision Public

    Dataset engineering on IP102 and benchmark of 8 detectors (YOLOv7-v11, SSD, Faster R-CNN); YOLOv11s exported to TFLite

    Jupyter Notebook 1

  4. docx-certificate-generator docx-certificate-generator Public

    Desktop tool that generates one filled Word certificate per person from a list (python-docx, Tkinter)

    Python 1

  5. event-ticketing-postgres event-ticketing-postgres Public

    Event ticketing database: PL/pgSQL functions, Flyway migrations, Jenkins CI/CD, SQL tests

    PLpgSQL

  6. pest-detection-system pest-detection-system Public

    End-to-end mobile pest detection: YOLOv11s (93.8% mAP@50) on-device via TFLite, Android app + Django API