Machine Learning Engineer · PyTorch · TensorFlow · ONNX · LLM agents
Dar es Salaam, Tanzania · UTC+3 · Remote worldwide or relocating
Tip
Hiring? I'm open to remote ML / AI engineering roles worldwide — including ML-heavy backend work — and to relocation. Contract or internship now, full-time from October 2026. EAT (UTC+3) covers the full European workday and US East Coast mornings. Fastest reply: isaackjoshua23@gmail.com
I'm a machine learning engineer who builds AI systems end to end: training and fine-tuning models in PyTorch and TensorFlow, exporting them to ONNX so inference runs on the device people actually have, and building the FastAPI services and apps that carry them into production. Most of my work is medical imaging for low-resource clinics — offline, on-device, low-bandwidth.
class Isaack:
role = "Machine Learning Engineer"
now = "ML Intern @ ETH Lab, MUHAS"
education = "B.Sc. (Hons) Computer Science, St. Joseph University in Tanzania (2026)"
based_in = "Dar es Salaam, Tanzania · UTC+3"
speaks = ["English", "Swahili"]
focus = ["computer vision", "medical imaging", "on-device inference", "LLM agents"]
also = ["FastAPI backends", "model serving", "Linux deployment"]
open_to = ["remote (worldwide)", "relocation", "full-time", "contract", "internship-to-hire"]
def principle(self) -> str:
return "Models should ship, not just score."Machine Learning Intern — ETH Lab (Emerging Technologies for Healthcare), MUHAS · Mar 2025 – present
- Built Mwana AI, offline breast-ultrasound classification for Tanzanian clinics — PyTorch → ONNX Runtime → Flutter
- Trained a CNN for TB/HIV co-infection detection from chest X-rays, with Grad-CAM explanations and ROC/AUC evaluation across demographic subgroups
- Fine-tuning RETFound for diabetic retinopathy; transfer learning on cardiac imaging for dilated-cardiomyopathy prediction
- Structured data for a multi-institution respiratory study (Aga Khan University, University of Warwick, NTLP)
B.Sc. (Hons) Computer Science — St. Joseph University in Tanzania · 2023 – 2026
|
ML · ON-DEVICE BREAST ULTRASOUND SCREENING Classifies lesions into BI-RADS categories, segments the region of interest and writes a structured clinical report — fully on-device, no cloud after first setup. Built at the ETH Lab (MUHAS) for clinics where radiologists are scarce.
Case study → · Android + iOS · source private |
ML · CHEST X-RAY TUBERCULOSIS SCREENING A CNN that classifies chest X-rays as Normal or Tuberculosis, trained on the public TB Chest X-ray dataset. Ships with a reproducible training script, a
|
|
ML · DISEASE-OUTBREAK EARLY WARNING A modular platform for predicting disease outbreaks in Tanzania, designed to take in health-facility, climate and mobility data and surface early-warning signals for public-health teams. Plug-in data sources let it grow across regions and diseases without rewriting the core pipeline.
|
AI AGENT · MACHINE DIAGNOSTICS Point it at a faulty computer: it runs tiered diagnostics, applies the software fixes it safely can, and escalates hardware or high-risk issues to a human. What it's allowed to touch is enforced in the architecture, not left to the model.
|
|
LLM APP · AI TUTOR FOR THE DESKTOP Turns any LLM into a structured, interactive tutor — explanation, worked examples and adaptive questioning instead of one-shot answers. Bring your own key and model (Claude, GPT, Kimi or local): no vendor lock-in.
|
BACKEND · MOBILE-MONEY ESCROW FOR EAST AFRICA Holds funds until the buyer confirms delivery. Selcom mobile-money deposits and payouts (M-Pesa, Tigo Pesa, Airtel Money), SMS OTP on every financial action, atomic state transitions, an immutable audit log and HMAC-verified webhooks.
|
| Project | What it is | Stack |
|---|---|---|
| Image_info_Extractor | CLI that reads echocardiogram report pages with Claude vision and writes one structured row per patient to Excel | Python · Claude API |
| muungano-bot | Swahili WhatsApp bot that answers questions about the Union of Tanganyika and Zanzibar from a vetted knowledge base — built for a university innovation contest | Python · FastAPI · Twilio · Gemini |
| DICOM → Video | Desktop tool that turns multi-frame, multi-series DICOM studies into MP4 / AVI / MKV with correct windowing and preserved frame rate | Python · PyQt6 · pydicom · FFmpeg |
| Izy | Local-first focus companion — an always-on-top mascot that tracks whether you're on task and holds natural-language reminders | Python |
| InviteFlow | Event invitation management with QR-code check-in and Twilio messaging | React · Express · PostgreSQL · Twilio |
|
ML & deep learning Training, fine-tuning and export — PyTorch, TensorFlow/Keras and Hugging Face through to ONNX Runtime, so inference runs on the device instead of a server. Computer vision for medical imaging: classification, segmentation, Grad-CAM.
|
LLMs & model serving LLM tool-use agents with hard safety boundaries and vendor-agnostic LLM apps. FastAPI services put models behind an API, with Celery and Redis keeping slow work off the request path.
|
Ship & run Ubuntu servers with nginx, Docker and TLS. Flutter for mobile, Electron and PyQt6 for desktop. Offline-first wherever the network can't be trusted.
|
Hiring for ML / AI engineering — or backend work that sits close to the models? Full-time, contract or internship-to-hire, email is the fastest way to reach me. Tell me the constraint you're building against and I'll tell you straight whether I'm the right person for it.
PROJECT.EXE exited with status 0 · built in Dar es Salaam, Tanzania