Mental Health Support Chatbot
A conversational AI project built around the difficult part, the data.
Collected and cleaned conversation examples from public sources, fine tuned GPT 2 and DialoGPT models, combined generative responses with Rasa workflows, and deployed the result as a web chat application and REST API on Google Cloud Platform.
The project explored how a language model could support a structured conversation rather than simply continue a text prompt. The data had to be collected, cleaned, and connected to a dialogue flow before the model could be useful.
Preparing the conversations
The collection pipeline gathered examples from public websites, forums, Reddit, and support communities. Cleaning, normalization, and deduplication were necessary before the text could become training material.
Combining two kinds of control
Rasa handled structured conversation paths while the fine tuned language models provided generated responses. The deployed application exposed that combination through a chat interface and an API.
selected tools
- Python
- Hugging Face
- GPT 2
- DialoGPT
- Rasa
- Flask
- GCP