The below blog is put together by Tanvi from Myna Mahila Foundation (tanvi@mynafoundation.com)
About the org and the use-case
Myna Mahila Foundation (https://mynamahila.com/ ) , a non-profit social enterprise operating in India’s urban slums, focuses on empowering women and girls through three core initiatives: Myna Health, Myna Employ, and Myna Research. These programs provide access to menstrual hygiene products, employment opportunities, and health education.
In this specific use case, we are developing Myna Bolo, a chatbot that harnesses the capabilities of Large Language Models (LLMs) with advanced features in prompt engineering and context augmentation. Myna Bolo is designed to offer non-judgmental, confidential, and medically accurate guidance, along with actionable advice on family planning for women in Indian slum communities.
Top Takeaways
- In our pursuit of advanced conversational AI, we’ve chosen to prioritize ChatGPT 4 over other LLM to have the first prototype
- While open-source LLM options are available, they may not be the best choice due to high hardware costs and potential quality gaps compared to GPT 4.
- An essential element is crafting system prompts in a way that yields the best response quality.
- The importance of a well-defined evaluation metric for assessing progress effectively in this endeavor.
Prototyping Done
- We’ve undertaken the task of generating system prompts, an essential step in the process.
- Additionally, we’re actively involved in the formulation of clear and effective evaluation metrics.
Potential Next steps / Help needed
- We are currently focused on integrating the model with WhatsApp to enable natural conversational responses.
- Additionally, we are dedicated to maximizing the utilization of our training data to make the bot more locally attuned in terms of its vocabulary and conversational style.
Overall thoughts on the sprint
Positive: The sprint was exceptionally well executed, from the planning to seamless coordination and execution. The best part was how time seemed to fly by; I was fully engaged throughout the two days. Everyone I met was incredibly welcoming, readily sharing their knowledge. It felt like we had known each other for a long time. Most importantly, it left me with valuable insights and a strong desire to collaborate more with the cohort. I eagerly anticipate more sprints in the future.
Can be improved: Having a brief introduction with the names and a two-line bio of all the attendees, including the host organization, would enhance the experience. This would make it easier to connect with others during the sprint, whether for assistance or simply striking up a conversation.