Open to opportunities

Joseph Nkumah

@josephnkumah

Message

AI/ML engineer deploying fine-tuned, quantized LLMs with cost-aware RAG and edge/offline inference.

Nigeria

Message

What I'm looking for

I’m looking to own LLM systems end-to-end—fine-tuning to deployment—building cost-aware RAG and reliable inference gateways, including quantized models for constrained edge or offline hardware with production monitoring.

I’ve spent 3+ years across the full LLM lifecycle—fine-tuning with Hugging Face (PEFT, LoRA, QLoRA), then moving models into production with quantization, inference serving, and monitoring. At NanoMatriX International Ltd., I deployed Hugging Face TGI and vLLM on AWS EC2 GPUs, built FastAPI inference gateways with async streaming and autoscaling, and used GPTQ to cut EC2 GPU costs while improving tokens-per-second throughput.

More recently, I founded ConfamNode with LiteLLM routing and async FastAPI/SSE on Jetson edge hardware, and shipped a quantized 4B medical LLM (MedGemma) to W4A16 in production on Jetson AGX Orin. I also co-architected a multi-tenant RAG system at SabiDocs using per-user isolation, Qdrant metadata filtering, and cost-aware inference routing.

Experience

Work history, roles, and key accomplishments

Current

Founder & AI Infrastructure Engineer

Current

JoTeq the First — ConfamNode

Jun 2026 - Present (2 months)

Founded and built ConfamNode, a sovereign, NDPA-compliant AI inference platform running on Jetson edge hardware. Quantized and deployed a 4B-parameter medical LLM to production on Jetson AGX Orin.

fastAPI LiteLLM Qdrant VLLM TEI (Tempo Entre Interações)PostgreSQL Redis Docker NVIDIA Jetson AGX Orin SSE

Current

Co-Founder & AI Engineer

Current

SabiDocs

Feb 2026 - Present (6 months)

Co-architected a multi-tenant RAG document-intelligence platform with per-user data isolation and cost-aware inference routing.

RAG Qdrant LiteLLM Redis PostgreSQL fastAPI

Current

AI/ML Engineer & Backend Infrastructure

Current

NanoMatriX International Ltd.

Oct 2023 - Present (2 years 10 months)

Fine-tuned and adapted LLMs using Hugging Face with PEFT, LoRA, and QLoRA. Deployed high-performance inference services with Hugging Face TGI and vLLM on AWS EC2 GPU instances.

Hugging Face PEFT LoRAs QLoRA VLLM TGI EC2 fastAPI MLFlow GPTQ GGUF Ollama Jetson Orin Nano NVIDIA Jetson AGX Orin

Machine Learning Engineer Fellow

Fellowship.ai

Apr 2023 - Jun 2023 (2 months)

Researched a state-of-the-art fashion image modification model using advanced deep learning techniques. Implemented computer vision models including HR-VITON, GP-VTON, and DINOv2.

Deep Learning Computer Vision HR VITON GP VTON DinoV2

Web Developer

Luday

Mar 2021 - Dec 2021 (9 months)

Built dynamic, cross-platform, responsive front-end websites using HTML, CSS, Bootstrap, and JavaScript. Delivered the company website used by clients to evaluate services online.

HTML CSS Bootstrap JavaScript Agile

Software Engineer Intern

NanoMatriX International Ltd.

Jan 2020 - Apr 2020 (3 months)

Built interactive UI components for mobile and desktop users. Collaborated with designers and sales teams to meet content and functional requirements.

UI Components Documentation Performance Testing

Education

Degrees, certifications, and relevant coursework

Liverpool Hope University

Master of Science, Advanced Computer Science

2022 - 2023

Grade: Distinction

MSc in Advanced Computer Science with Distinction, covering Artificial Intelligence, Big Data, IoT, and Data Analytics.

Bingham University

Bachelor of Science, Computer Science

2011 - 2015

Grade: Second Class Upper

BSc in Computer Science with Second Class Upper, covering Web Development, Data Structures & Algorithms, OOP, Database Management Systems, and Artificial Intelligence.