SafeGPT

• 

  Jupyter Notebook Demo

Inspiration

I read a story of a woman in a domestic violence situation who managed to call 911 without the perpetrator realising, by pretending to order a pizza. The person who took the order could decipher the hidden message. Having someone on the other end of the call who understands you matter, so black people especially black women in distress/gender-based-violence situation needs first AI emergency call centre bot who understands their speak.

What it does

Safegpt is a Data-API service that enhances emergency call centers with inclusivity by capturing diverse voices. Leveraging advanced prompt engineering techniques and fine-tuning interfaces for LLMs. We empower call centers to better understand and serve marginalized communities, ensuring that everyone receives the support they need in times of crisis.

How we built it

For the prompt-engineering demo; We researched and curated systems prompt that makes a diverse, inclusive chabot We then use this prompt to make a better response from the most popular LLM, ChatGPT which is being used now by many call centres. The ChatBot programme was written in Python and leveraged OpenAI API

Challenges we ran into

Finding resources on call-centre logs that cater to marginalised, diverse populations. That we can clean and use to fine-tune LLMs,

Accomplishments that we're proud of

We are proud to implement an inclusive chatbot in about 24 hours.

What we learned

We learned about Bias in AI systems and how it disproportionately affects certain race and gender, prompt-engineering, fine-tuning, OpenAI API, Large Language Models

What's next for SafeGPT

1. Curating diverse speech datasets especially starting with African-American, Black, Hispanic, Native people.
2. The API interface system will enable call centres to fine-tune their speech data

Built With

• jupyter
• openai
• python