Field log — Abuja, Nigeria · N09.0765 ° E07.3986 °

Samuel
Edungbola

AI Engineer — LLM Systems, Retrieval & Full‑Stack ML

I design and ship the whole system, not just the model — fine‑tuning, RAG, and computer vision, wired into real Next.js and FastAPI products running in production.

4+ yrs in the field · Abuja → London → New York (remote)

Follow the route

01Current position

Technical Operations Manager — de facto CTO

Prime Frontier Group

Sep 2025 – PresentRemote (New York)

  • Sole owner of technical strategy and architecture across Prime Frontier Group and every subsidiary — the full stack of decisions, from platform delivery down to which cloud services get used.
  • Built and lead SAMIS, an ML‑driven mineral‑prospectivity model piloting in Peru, fusing geospatial data with predictive modelling for exploration targeting.
  • Shipped AI chatbots for internal and client‑facing workflows, and drafted a proposal for an LPR + real‑time object‑detection vehicle‑access security system.
  • Re‑architected cloud spend across AWS and Vercel, cutting external infrastructure cost by 80%.
  • Also architect of record for eleven further group platforms — drilling exploration (live), renewable energy, procurement, accounting, and a 41‑person company intranet built end‑to‑end in‑house.

02Consulting engagements

AI Engineer Consultant

Kouture Konnect

Oct 2025 – PresentConsultancy · Remote

  • Built a production RAG pipeline — LangGraph, GPT‑4o Nano, text‑embedding‑3‑large, Chroma — grounding a fashion retailer's chatbot in its own live catalogue instead of generic model knowledge.
  • Recommendation engines for product, design, and stylist discovery, fed by conversion signals and catalogue metadata.
  • Daily automated ingestion pipeline keeping the AI knowledge base in step with an evolving catalogue.
NDCG
0.79
MRR
0.74
Keyword coverage
91%

AI Evaluation Consultant

Alphabet (via Invincible)

Oct 2025 – Nov 2025Remote

  • Rubric‑based evaluation of autonomous agents on complex multi‑step tasks — task completion, factual accuracy, reasoning efficiency.
  • Reviewed full action trajectories and artefacts to catch hallucinations and reasoning gaps, feeding structured findings back into model and rubric improvements.

03Independent contractor · concurrent with Prime Frontier Group

Senior Machine Learning Engineer

Pingtop / GreenApps (Tezda)

Jun 2025 – Jan 2026Remote (London)

  • Architected the ML systems behind a growth curve of 100,000 users in 2 months and 1,000,000 in 5, sustaining 80,000 daily actives.
  • Designed a slot‑based feed‑distribution engine — trending, new, following, and item‑based slots — A/B tested for per‑session retention.
  • Built a multimodal moderation pipeline (Falcon NSFW, BERT, Whisper) on AWS EKS at ~99% accuracy with near‑zero false negatives.
  • Fine‑tuned Gemini 2.5 Flash for moderation and for video description/keyword extraction, powering a full video search engine.

Machine Learning Engineer

Mar 2025 – Jun 2025Remote (London)

  • Built Tezda's virality‑prediction engine from TikTok, YouTube, and Instagram signals — 10,000+ videos, early‑phase features against late‑phase viral labels.
  • Neo4j‑backed mutual‑friend and influencer fallback recommendations.

04Foundations

Health Informatics Assistant

Heartland Alliance Nigeria

Nov 2024 – Feb 2025Abuja, Nigeria

  • Led EMR database administration and data‑integrity coordination across 9+ facilities on the KP‑CARE 1 project.
  • Power BI dashboards lifted data concurrence 12% and module‑version alignment 45% across 19 facilities.

Backend Engineer

LifeCare

Mar 2022 – Apr 2024Remote

  • Designed the company's core database systems, including access controls guarding personally identifiable information.
  • Built the company website and the general operational APIs the business ran on.

05Featured builds

Shipped, not benchmarked

CHIOMA

Whisper · Deepgram · LLM + RAG · ElevenLabs · Cloudflare Workers

Voice‑first customer‑support AI handling calls in English, Pidgin, Yoruba, Igbo, and Hausa — VAD → STT → LLM+RAG → TTS, targeting sub‑500ms response latency.

chioma-fe.sedungbola.workers.dev

GPT‑4.1 Fine‑Tuning

PyTorch · Hugging Face · QLoRA · PEFT · LoRA

4‑bit quantised domain adaptation that preserves general world knowledge, with PEFT/LoRA adapters enabling versioned, low‑cost client deployments.

Football Match Prediction Benchmark

XGBoost · PyTorch · scikit‑learn · FastAPI

Leakage‑safe features and Platt/isotonic calibration across tabular ML, a PyTorch MLP, and a fine‑tuned LLM — judged on calibrated betting ROI, not accuracy alone.

SabiArtisan

Next.js · FastAPI · PostgreSQL · Computer Vision

Verified artisan marketplace for Nigeria: a 3‑tier trust engine (phone, ID, live snapshot), escrow protection, and competitive bidding.

sabiartisan.com

06Systems & tools

Working materials

LLM & GenAI
QLoRA / PEFT / LoRA fine‑tuning (GPT‑4.1, Gemini 2.5 Flash) · LangGraph · RAG · GPT‑4o · hallucination evaluation
ML & Vision
PyTorch, TensorFlow, scikit‑learn, XGBoost · object detection, emotion & face recognition
NLP
spaCy, BERT, OpenAI Whisper, semantic embeddings & text similarity
MLOps & Infra
Docker · AWS (EKS, SQS, Personalize, Rekognition, Comprehend) · CI/CD · model monitoring
Vector DBs
Chroma, FAISS · semantic search & similarity retrieval
Full‑Stack
Next.js, TypeScript, Tailwind CSS · FastAPI, Flask · PostgreSQL, MongoDB, Neo4j

07End of route

Let's build
the next one.

Open to AI engineering roles where shipping the full system — model, pipeline, and product — is the job, not a side quest.

sedungbola@gmail.com +234 911 5953 083 linkedin.com/in/samuel‑edungbola github.com/sedungbola

Off the clock: rapid chess, ranked 98th percentile globally.