Princewill Inyang Available for backend and AI roles
Princewill Inyang · Backend and AI engineer

It usually starts with one sentence.

Heard at work

I hear the sentence, work out what the business actually needs, and ship it: AI agents, ERPs, telehealth platforms and the data pipelines underneath them.

See it working Email me
Portrait of Princewill Inyang
Princewill InyangPython · TypeScript · AI systems
2 months

From a spreadsheet prototype to a system every department uses.

17→19

Odoo ERP upgrade across two major versions, live on the first attempt.

8→1

Repositories merged into one tested codebase for a US telehealth platform.

2 brands

Telehealth brands live on one platform, the second without a rebuild.

Selected work

Seven sentences, and what I shipped for each.

01AI task intelligence

Nexus

Vendease · B2B food supply, Lagos and Abuja

“Tasks keep getting lost.”

The problem
Asks arrived in Google Chat, email, docs and meetings, and some quietly vanished.
What I built
An AI system that picks up each ask wherever it was made, gives it an owner, tracks it to a decision and flags anyone carrying too much.
The result
From a spreadsheet prototype to company-wide use in two months.
PythonLLMsGoogle Workspace APIs
Replaysample data
02WhatsApp commerce

Koko

Discover Fresh by Vendease

“Restaurants want to order where they already talk.”

The problem
Customers lived in WhatsApp. Ordering meant leaving it.
What I built
A WhatsApp agent where restaurants browse, build a basket and pay by Paystack link without leaving the chat. Paid orders are locked against double charging, and staff can step into any live chat, mute the agent and hand back.
The result
Live since July 2026, tested against the real model, not a stand-in.
FastAPILangGraphGeminiPaystackWhatsApp Cloud API
Replaysample data
03ERP

Odoo, two versions up

Vendease

“Move us to the new version without stopping the business.”

The problem
The ERP ran finance, pricing and delivery, and it was two major versions behind.
What I built
I learned Odoo on the job, gave finance a credit line for customers, a pricing engine with audited discount ceilings and a 13-week cash forecast, then ported ten custom modules from 17 to 19, rehearsed on a copy of the live system. When the old platform was about to be switched off, I rescued its data in time and turned it into a weekly list of past customers worth winning back.
The result
Live on the first attempt, with minimal downtime.
OdooPythonPostgreSQLpgvector
Replaycutover log, simplified
04AI patient assistant

Alfie

ProxyMedicine · US telehealth

“Patients need help between appointments.”

The problem
A simple triage chatbot couldn't handle refills, lab orders or bookings.
What I built
A multi-agent assistant. Every message is checked for emergencies before any model runs. Agents ask for forms, choices and booking cards through typed tool calls streamed to the app, and each one reaches the clinical platform over MCP with only the tools it needs.
The result
The assistant patients use for triage, refills, lab orders and bookings.
LangGraphRAGpgvectorMCPSSEReact
Replaysample data
05Telehealth platform

The ProxyMedicine platform

Main backend engineer

“We want a second brand without building a second platform.”

The problem
The clinic's code was split across eight repositories, so every change was slow and risky.
What I built
Onboarding, ID verification, triage, scheduling, Stripe payments and e-prescribing. I merged the eight repositories into one tested codebase, and built the analytics that show what the clinic earns and what each new patient costs to win.
The result
A client's second telehealth brand, StatWellMD, went live on the same platform without a rebuild.
FastAPIPostgreSQLStripeTwilioAWS TextractDoseSpot
Replayarchitecture, simplified
06AI sales research

Lead sourcing

Demz Aminytics · health e-commerce

“Find us more customers like our best ones.”

The problem
Qualifying a health brand meant someone reading its whole website.
What I built
Gemini agents read a company's site the way an analyst would and score it out of 100, with the evidence kept. Then the system takes the top scorers, writes a dozen targeted searches for each, and checks every result seven ways before it becomes a lead.
The result
A lead list that grows from its own best results.
PythonGeminiFirecrawlMongoDB
Replaysample data
07Music analytics

Chartex, version one

Built with the founder

“Which songs are spreading on TikTok before the charts notice?”

The problem
By the time a song charts, the moment to act on it has passed.
What I built
The first version of what became Chartex: a backend that records how many TikTok videos use each sound every day, turns that into growth, discovers new sounds from creators' profiles and feeds song recognition.
The result
Chartex grew into a music analytics platform used by record labels.
DjangoFastAPIPostgreSQLDockerPlaywright
Replaysample data
How I work
I test the request against real data before I build it.

Rules that look right on paper are often wrong in practice. Checking first means they never reach production.

Also built

Before the sentences got bigger.

2022 to 2024

Creative analyticsDecoditive

Took over the company's computer-vision services, then built the next generation: an image-analysis API that scores every ad creative on colour, contrast, objects, faces and on-image text, and in 2023 an AI ad generator that writes the copy, creates the visuals and lays out on-brand ads.

Audience data

Social data pipelinesDecoditive

Pipelines that collect public posts and comments for a watchlist of accounts and profile the audience behind them with computer vision, multilingual sentiment and LLMs, delivered on a schedule in one consistent format.

Early learning

BloomieDemz Aminytics

The daily lesson platform behind Bloomie, an app for children aged three to six: five age-matched video lessons a day, quizzes read aloud, a retry queue and progress for parents. Bloomie is live at bloomie.com.

Health coaching

AlexBoldBetterLiving

A coaching agent with Gemini vision body-scan analysis. Traces showed a 19,000-token system prompt slowing every turn, so I cut it down.

Writing

I write down what I learn.

Semaphore
RAG-enabled LLMs for data analysis
Read Semaphore
Custom middleware in FastAPI
Read Semaphore
Docker volumes
Read Earthly
Linux networking commands
Read MakeUseOf
Creating and distributing Python packages
Read Blog
More on my blog
Visit
Toolbox

Whatever the problem calls for.

Languages

Python, TypeScript, JavaScript, SQL, Bash

Backend

FastAPI, Django, Flask, Node.js, Express, Odoo, REST, SSE, webhooks

AI

LangGraph, LangChain, MCP, Pydantic AI, RAG, LangSmith, Gemini and OpenAI APIs, OpenCV

Data

PostgreSQL, pgvector, MongoDB, Redis, pandas

Infrastructure

Docker, GitHub Actions, AWS, RabbitMQ, nginx, Linux, Playwright

Integrations

Stripe, Paystack, Twilio, DoseSpot, WhatsApp, Google Workspace

Got a sentence like that?

princewillingoo@gmail.com

Open to backend and AI engineering roles and available immediately.