Portfolio
The work,
end to end.
Production AI and backend systems I've built for teams in London, the Gulf and Southeast Asia, from Lagos. Some clients are public, others stay confidential, so a few of these are described without the name attached.
- 9
- Products shipped
- 23K+
- Users reached
- 4
- Continents
Selected Work
Everything I've shipped.
Pocket Coach · D Jungle People
Backend and AI for a voice-first leadership coach, live on iOS and Android, built for D Jungle People, a respected 20-year Southeast Asian coaching firm, and modelled on the founder's own coaching cadence. An LLM runs a strict T-GROW protocol from their proprietary frameworks: one question per turn, exploration before advice, and a confidence-scored close, with speech-to-text in, neural text-to-speech out, and vector retrieval over the coaching corpus. It reached £100k+ in new revenue within four months of launch, with a 20% lift in post-workshop retention.
Backend & AI
2025
Shyft
Contributed to the backend and AI behind Shyft, a cross-platform NHS healthcare-staffing ecosystem of four connected products: a mobile app for professionals plus internal, agency and hospital dashboards, all on one shared backend. Worked on the real-time data sync, automated compliance tracking and smart job-matching that connect 1,492+ medical professionals to 4,990+ jobs at an 87% satisfaction rate.
Backend & AI
2026 – present
Apprentago
AI-powered apprenticeship platform. I built the complete backend and API layer for 22,000+ users and 10,000+ job listings, powering multiple dashboards for students and employers on one central backend.
Backend & AI
2024 – 2025
Memory Lane Games
I built the AI voice companion for Memory Lane Games, an established dementia-care app, adding an empathic conversational layer to their existing product. It reads emotional prosody in the voice, not just the words, and paces itself for elderly listeners. When someone reminisces it tool-calls into a photo memory quiz, read aloud and scored with error-less learning that never says the word wrong. Distress cues escalate to a human carer, and the whole behaviour is bound to a clinical safeguarding spec drawn from Validation Therapy.
AI & Backend
2025
ForeSight
As one of a two-person team, I built ForeSight full-stack, frontend, backend and AI: R&D tax report-writing software and CRM. I owned it end to end, from the data model and API layer through to the interface, with an LLM drafting the technical R&D tax reports at its core.
Full-stack · AI, Backend & Frontend
2026 – present
Snugs Baby
An AI weaning assistant for breastfeeding and early-stage mothers, built full-stack into the brand's existing Shopify store. Runs on OpenAI, grounded in NHS and WHO guidance via a RAG pipeline (FAISS + embeddings), with per-baby profiles and age-aware advice, voice, photo and document input, automatic safety checks on every response, food-introduction tracking and matched product recommendations from the Shopify catalogue.
Full-stack · AI, Backend & Frontend
2025
XCARE Capital
Core member of the two-person backend & blockchain team. Built the self-custodial wallet infrastructure from inception, integrating Aave, Morpho, Kamino & Jito with real-time position tracking across Ethereum & Solana, plus in-app calls and messaging.
Backend & Blockchain
2025 – presentArabic–English Legal Chatbot
A bilingual English and Arabic legal assistant, built full-stack for a client with strict data-sovereignty needs, handling 200+ queries a day. Llama 3 on AWS Bedrock answers questions, grounded in documents users upload through a LlamaIndex RAG pipeline, with on-the-fly translation and a fully right-to-left Arabic interface. Owned it end to end: Django backend, the retrieval and language pipeline, and the React frontend.
Full-stack · AI, Backend & Frontend
2025Engineering AI Assistant
I took this from an exhibition prototype to a working product, built full-stack: an AI assistant for a building-services engineering firm. Engineers upload PDF drawings for entity recognition and CAD-readable conversion, then ask questions answered from the CIBSE guides and the client's own documentation. I built the retrieval on an in-process FAISS index of Titan embeddings, with Llama 3 on AWS Bedrock answering only from retrieved context and citing the source guide and page.
Full-stack · AI, Backend & Frontend
2025A note on confidentiality
Some of the best work stays quiet.
Client engagements come with NDAs, so a few systems here are described by what they do rather than who they were for. Where an employer has published a case study naming the client, I link straight to it. Happy to talk through the architecture and trade-offs of any of them in a conversation.
On the full-stack builds I own the AI and backend end to end, then build the interface myself, driving coding assistants like Claude Code and Cursor to translate the Figma designs into working UI. It keeps the delivery fast without handing off the parts that matter: the architecture, the model behaviour and the API contract stay mine.
Let's talk
Let's build something ambitious.
Whether it's an AI system, a backend platform or a research collaboration, let's start a conversation.