Power to Change’s Discovery Fund and programme was aimed at community businesses looking to explore tech-based solutions to meet challenges faced by their community. 

In brief, we wanted to create (agile) solutions to the communication problems we regularly experience with the young people we work with. Or, to flip it on its head - what could we do to make communicating with our communities work better? This addresses multiple needs: frictionless (or less friction-filled) comms means that we can communicate out into our community in a way it will be more likely to be received. And on the other side of the equation, we are looking at how young people can communicate in to us. Both have the same goal - to engage (and retain) disadvantaged young people wanting to secure better (life/ work) outcomes for themselves through the training and support that we (and potentially in future - others) provide.

We also branched out to not just include comms in and out, but communications in terms of what type of content people receive and when whilst undertaking learning with us.

Summary of tech explored during the Discovery Programme:

• We explored LLMs for automated code review, as well as supporting tech for realtime/group collaboration. This fulfils several needs – giving immediate actionable feedback to learners as they get to grips with foundational concepts and syntax; and also freeing up our volunteer mentors for “higher quality” advice and support (e.g. coaching, mentoring, and chat rather than mechanical “your brackets are in the wrong place” guidance)
• We also looked at multiple community tools and services before settling on “all-in-one community platform” Circle. We have also earmarked some purely OSS solutions to explore further in the future - including live community video (for e.g. conferencing, 1:1 mentoring, meetups, online workshops, etc.) which are directly integrated into our own platform (via Livekit)
• We prototyped and built a prototype learning platform using Elixir and Phoenix - an OSS framework for building web tech with real-time functionality such as e.g. chat and video, live collaboration, etc. (and is the same underlying tech used by WhatsApp and Discord)
• We used Notion (and Miro) for project management although we have explored using the OS notes app Outline
• Language models. We took a foray into ChatGPT to test how it might work, on say automated code reviews. GPT-4 (and now -4o) is the most effective solution out there currently for code analysis (link), although we are hoping to see some strong OS candidate LLMs in the future
• Generally applicable (i.e. non-vendor-specific) language model techniques we’ve explored to date include RAG (retrieval augmented generation), and multi-step “inner monologue” prompting as a workaround to LLMs’ current issues with reasoning. Not fully explored currently but earmarked for future experimentation are the OS model testing (for safety, value alignment, etc.) framework from Giskard, as well as prompt benchmarking and evaluation via e.g. Langchain Evaluators

Read more about the case for community tech here