Oghenetega Biokoro

At HeartStamp, I built the generation Orchestrator and Inference Engine from the first architecture diagram through production alpha, routing across 12+ image, video, and language model providers. I also cut end-to-end card generation time from roughly two minutes to one through concurrent panel generation.

I build model evaluation systems that make production model selection measurable across quality, edit fidelity, facial consistency, multilingual rendering, latency, and cost. I’ve also created intent-based routing, provider health checks, fallback chains, multilingual image-editing workflows, and a model-agnostic image-to-video integration.

At cozy.art, I built the inference server behind generative model requests, including GPU memory management, runtime optimisation, quantised model variants, LoRA pipelines, and RunPod-based orchestration. I contributed AuraFlow architecture support to Hugging Face Diffusers and bring five years of Python experience, including three years shipping ML systems remotely.