AI Product Manager. Filmmaker.Both, on purpose.

I build AI products from the model to the frontend, with explainability and ethics built in. I also direct AI commercials and short films using the same systems. Same craft. Different output.

Two paths. One person. Pick the one you came for.

About Me

Ogbebor Osaheni

AI Product Manager

I’m Ogbebor Osaheni, an AI Product Manager building end-to-end ML products from the ground up: ML pipelines, FastAPI backends, and Next.js frontends shipped to production with full PM artefacts before a single line of code.

I’m drawn to products where AI meets human creativity and editorial judgement. Bias detection tools that keep reviewers in control, story engines that guide writers rather than replace them, pricing systems that explain every recommendation in plain English. In each one, the AI shows its reasoning instead of asking to be trusted on faith.

Open to AI PM roles in the UK, particularly in media, film and creative technology. Projects built using AI-assisted development workflows, directing Claude Code and v0 to ship production frontends while maintaining full ownership of product decisions, architecture, and PM artefacts.

ML & XGBoost

Built XGBoost pipelines across pricing, sentiment, and bias detection. Applied Optuna TPE search (50 trials, 3-fold CV) on the pricing system — confirming the baseline was already well-calibrated and the R² ceiling was genuine market noise, not fixable by tuning.

Explainable AI

Applied SHAP to the pricing and bias detection models. Pricing: 3 chart types (bar, dot, waterfall) with live per-prediction explanations. Bias Audit: per-word SHAP waterfall in every result card — reviewers see exactly which words drove the verdict and by how much.

Product Strategy

Wrote PRDs, OKRs, and roadmaps before building. Shipped PM artefacts first, code second — on every project.

Responsible AI

Built bias audits, fairness constraints, and ethics frameworks aligned to UK Online Safety Act, Ofcom, and GDPR.

Full-Stack Shipping

FastAPI backends, Next.js frontends, Supabase — from model to deployed product, end to end.

AI-Assisted Development

Direct Claude Code and v0 to ship production interfaces, maintaining full ownership of decisions and architecture.

Projects

A selection of my AI product management and backend development work.

AI Dynamic Ticket Pricing

Python

XGBoost

SHAP

An end-to-end AI pricing system for UK live events that integrates artist popularity, venue location premiums, demand urgency signals, and market conditions to generate explainable, ethical price recommendations

Case Study Demo

Frame Intelligence

Next.js

FastAPI

Claude API

AI-powered commercial film intelligence platform for professional filmmakers and creative directors. ATANDA Lens analyses TikTok and Instagram commercials across 6 layers (Brand, Story, Cinematography, Sound, Post, and Performance) using Claude API and computer vision with Bruce Block and SB7 frameworks. ATANDA Forge delivers a 7-stage AI production pipeline from brief to post-production brief, powered by the Meridian Engine.

Case Study Demo

Bias Audit Dashboard

Next.js

FastAPI

XGBoost

B2B AI tool for trust and safety teams at UK media companies. Detects bias across 6 categories using TF-IDF + XGBoost (F1 0.90) with tiered routing: short inputs go to Claude API for semantic classification, long inputs to the full XGBoost + SHAP pipeline. Includes a live comment moderation simulator, fairness metrics panel, and audit log. Built for Online Safety Act 2023 and Ofcom compliance.

Case Study Demo

TrueCase

Next.js

Claude API

TypeScript

AI-powered business case builder for AI investments. Generates governance-adjusted ROI projections with a reliability score (0–100%) based on six KB-backed governance elements. Claude Sonnet 4 produces a three-section business case narrative constrained entirely to a verified knowledge base, not training data. Zero data retained. Exports board-ready PDF and HTML.

Case Study Demo

Pulse: AI Audience Sentiment Monitor

Python

FastAPI

Next.js

Real-time AI audience sentiment monitor for live UK broadcast events. TF-IDF + XGBoost emotion classifier (5-class, Macro F1 0.830) and multi-label topic classifier built for live broadcast producers at BBC, Channel 4, ITV, and Sky. Scripted BAFTA simulation with narrative arc, negative sentiment spike alerts, SHAP word-level explainability, and an editorial guardrail built into the architecture. Aligned with Ofcom Broadcasting Code.

Case Study Demo

Technical Fundamentals

Deliberate, hands-on practice with core ML and data workflows — documented honestly, mistakes included.

Practice Project — Not Shipped

Titanic Survival Prediction

A guided, hands-on ML fundamentals exercise: data cleaning, confound investigation, feature engineering, model comparison, and honest train/validation/test evaluation. Built to strengthen the technical judgment behind AI product decisions.

81.0% final test accuracy — +19.5 pts over lazy baseline

Read the Case Study

Practice Project — Not Shipped

Loan Approval: Bias & Compliance Audit

Building a model for a regulated domain from day one — compliance and fairness questions shaping every decision, not bolted on at the end.

82.9% test accuracy — but an honest, inconclusive fairness finding kept it from a Go recommendation

Read the Case Study

Learning Project — Not Shipped

Loan Policy Assistant

An agentic RAG assistant for loan officers — retrieval, tool use, guardrails, and evals, built to pair with the Loan Approval bias audit project and demonstrate governed AI product design.

0 guardrail failures — 14/15 eval pass rate across policy, prediction, and refusal questions

Read the Case Study

Learning Project — Not Shipped

Model Maintenance: Continuous Monitoring

What happens after a model ships — drift detection, retraining triggers, and a CI/CD pipeline designed to keep a production model healthy over time.

KS test drift gate, three-tier decision framework, real cross-project pipeline dependency

Read the Case Study

Get In Touch

Open to AI PM roles, collaborations, and interesting conversations about responsible AI in UK media.

linkedin.com/in/osaheniogbebor

Opens in new tab

Based in Salford, Manchester. Available for remote and on-site roles across the UK.