=== Hindustan Consultancy Services Mental Health Analyzer built on OpenAI gpt4o-mini === Filetree . ├── app.py # the flask app used for user input ├── db.py # script for initializing sqlite3 db ├── gpt.py # scripts for chain of thought and few shot prompting gpt4o-mini ├── inference.py # inference of the older llama3.2 and sentence transformers based models ├── pyrightconfig.json ├── readme.txt # the file you are currently reading ├── sheet.tsv # sample spreadsheet converted to tab seperated values ├── static │ └── grid.svg ├── templates │ └── home.html ├── train.py # contains the fine-tune loop for older concern classification model └── users.db # db file populated on running the server Method - we first used a sentence transformer model and fine tuned it on classification task (+ used a llama3.2-3b Instruct model for other categories) using the spreadsheet provided. It was only able to correctly answer one single concern which is why we moved on to prompt engineering a much powerful gpt4o model. - we are using chain of thought and few shot prompting on gpt-4o-mini. - provide it a step by step way to reason about the problem - ask it to return the answer in a structured 4-tuple (s1, s2, s3, s4) where s_i represents the i_th step of reasoning - we parse the structured tuple for displaying the results. - we pass the last five status updates with a prompt to use the 4-tuples for inferring the change in emotions over time. - a demo video has been uploaded to youtube: https://www.youtube.com/watch?v=7kAUg2A935M Instructions on getting it to run - get an openai key and add it to your environment first (you can add this to your ~/.bashrc) export OPENAI_API_KEY="sk-proj..." - create a venv (optional but recommended) python3 -m venv .whatever source .whatever/bin/activate - install dependencies pip install openai flask - run the flask server python3 app.py - open localhost:8123 on your browser === Akshit Kumar, Anushka Jain ===