I will audit and fix your ai model

Send me the repo and tell me what is wrong. I will give you my read before you order.

Your model reports 95% accuracy and still fails in production. I find out why.

6+ years of engineering experience, an Erasmus Mundus master's degree in AI, and 5 peer-reviewed AI papers. Designing evaluation and benchmarking for production AI is my day job. Evaluation is the part almost everyone skips.

What I usually find:

- Data leakage and broken train test splits

- Metrics that flatter, wrong averaging on imbalanced data

- Overfitting with no real validation

- Preprocessing mismatch between training and inference

- Slow inference, bad batching, wasted GPU

- For LLM and RAG: hallucination rate, weak retrieval, missing guardrails

You get a ranked report, working fixes with code, and an honest verdict. If the approach cannot work, I will say so and explain why.

Works on PyTorch, TensorFlow, scikit-learn, YOLO, Hugging Face, LangChain and RAG pipelines.