LLM Cohort at Kochi: Learnings, Use-cases and Resources from Experts


OCTOBER 26, 2023


Glific team started with the following objectives to host the 2 day offline sprint:

To bring the NGOs who are experimenting with LLM / AI technologies together on a common floor and enable 

  1. Level up in the understanding of the technology (through expert talks / showcases of the work) 
  2. Share the vision, experiments conducted, learnings and future requirements (through NGO showcases / talks) 
  3. Roadmap solutions for the same (through collaborative work time with experts and Glific developers)

Here are the top 5 takeaways from the talks shared by experts, reflections from NGOs and discussions over tea, late night and early morning walks

  1. Focus on the “value creation” or in other words “solving a problem” rather than implementing LLMs for the sake of it. 
  2. If you’re getting started, use the most easily, freely available resources at present (and there are plenty) instead of worrying about hosting an open source model, doing fine tuning specific to the sector and so on
  3. Do not worry about the cost at scale, when doing a pilot. The cost of technology is bound to come down in near future. And at present the cost of prototyping and piloting with the new developments is close to nil.
  4. Focus on prompt engineering for adapting LLMs to your usecase, that is where NGOs can now innovate and stand out
  5. Imagination is the one of the biggest barriers to how orgs can adapt LLMs, spaces where orgs can learn and share among peers is helping to open up the imagination. Most of what one can imagine is possible with the technology present.

About the NGOs who attended, their usecases and takeaways

5 NGOs got together at the sprint covering education, livelihood and healthcare sectors. 

Udhyam, are thinking of using Generative AI (like ChatGPT and other LLMs) to provide more personalised experience to the users, Improve the overall user engagement and learning, and elevate the content creation process. Here is a deep dive into the ideas as well takeaways from the sprint

Bandhu, Leveraging the power of LLMs to expand housing connections for 200 million low-income blue-collar workers. Read more on their usecase and work done so far here

Quest Alliance, leveraging the power of LLMs to provide personalized learning journeys based on local contexts such as previous learning experiences, geography or genders and so on. Read more of their ideation process, expectations and work done so far here 

Myna Mahila Foundation, leveraging LLMs to provide empathetic, non-judgemental and localised advice on Sexual and Reproductive Health to women from slums in Mumbai, read more on their usecase what they have put together as a proof of concept here

The Apprentice Project, leveraging LLMs the ongoing chatbot based program via TAP Buddy to get smarter and be capable of clearing incoming doubts regarding the content and weekly activities that imbibe social and emotional skills in the students on their bot, read more about the work done by them leveraging LLM here , here and here

Other emerging NGO use cases (who were not present for the sprint) for leveraging LLMs as a solution

Reap Benefit, imagining image recognition or description to identify the type of issue being reported byt the citizen via the image, aiding the story telling skills of a change-maker and providing mentoring tips and next steps to solve problems based on the help requested by the change-makers and the custom knowledge base of the change-making toolkits, read more on their use-cases here

INREM, imagining visual recognition capabilities to detect the pollutants in water like arsenic and fluorides, read more here

While most the NGOs are starting off experimenting with Question Answering over Custom Knowledge Bases, above are some of the more imaginative and important use cases

About the experts

Experts were shortlisted and invited based on their practical and hands-on experience of implementing projects for social sector as well as education based use-cases.

Above is Edmund Korley from Agency Fund who shared an insightful session from his learnings of implementing LLM powered solutions with grantees of Agency Fund in Africa. With a deep understanding of technology, passion for social impact as well as connects in the Silicon Valley, Edmund was able to provide a well rounded perspective and actionable direction on what the NGOs should focus on and what are the holes they should avoid falling in at the moment depending on their scale and use-cases. Go here to get the full scoop of Edmund’s session

Another prototyping expert we had was Aman Dalmia from Hyperverge Academy, who is building SensAI. Session from Aman extended over to 2 hours on popular demand and was of immense value to the NGOs to get the practical tips on what kind of mentality to have when starting off with LLMs, basics of technology and terminologies used, how to structure system prompts, what makes a good knowledge base, how to debug issues, how to break down the larger output expected from an LLM into smaller steps and loads more, refer here for the detailed dive into the session

A note of gratitude 

To the two experts Edmund from Agency Fund and Aman from Hyperverge Academy and their respective organizations who were generous with their time and knowledge. Every hour of their presence at the sprint over the 2 days was spent in conversations with NGOs, with Glific team clarifying doubts, adding to the development, reviewing the roadmaps, directing to solutions, suggesting the design of things and at times pondering the deeper philosophical questions related to LLMs.   

And finally a note of gratitude to CISCO, whose grant made this collaborative learning effort possible. Project Tech4Dev whose team members contributed with the logistical support and NGOs who participated with full enthusiasm to derive maximum learnings and forge deep connections with each other to carry forward from the space. 

Leave a Reply

%d bloggers like this: