looking for AI expert to fix chat bot urgent

My production chatbot has started missing the point of conversations—​it stalls or returns a fallback message whenever the user asks anything even mildly off-script. The root problem sits in its natural-language-understanding layer: intents are misclassified, entities aren’t picked up, and the confidence scores drop to zero far too often. The codebase is written in Python and currently relies on a transformer-based model served through a REST API, with a small Rasa NLU pipeline supervising intent detection. I’ve already confirmed the webhook and integration endpoints are healthy; this is squarely an NLU issue. What I need from you: • Inspect the NLU pipeline, training data, and model configuration. • Identify why the model fails to map typical user utterances to the correct intents. • Retrain or fine-tune the model (TensorFlow / PyTorch is fine—whichever you think best) and adjust confidence thresholds so responses trigger correctly. • Hand back an updated codebase, fresh training data (if added), and a short test script that proves at least five varied user queries are now handled accurately. Please keep the turnaround tight—​the bot is live on our site and every hour of misfires costs us conversions.

Project ID: 40421876