October 26-27 | Open Source in Energy Access Symposium Hackathon | Kigali, Rwanda
An open-source, NLP-powered chat assistant that guides users through the Energy Access Explorer (EAE).
Geospatial data plays a crucial role in integrated energy planning. EAE supports users with no or limited GIS programming or expertise to use EAE for energy planning by integrating granular data on both energy supply and demand, ensuring energy solutions are viable for service providers and end-users alike. EAE users are then able to visualize geospatial data, customize multi-criteria analyses, and identify and rank priority areas based on their perspective. However, not all users have a similar understanding of the tools, which often leads to not being able to utilize the platform effectively.
To bridge this gap, we are looking for a proposal for an open-source large language model, used through Natural Language Processing (NLP), as an interactive chat assistant. This system will guide users through complex tasks like creating detailed priority maps for potential electricity interventions (e.g. stand-alone PV, minigrid, existing grid) and comprehensive reports by analyzing datasets, including:
- Demographic information (e.g. population)
- Social and productive energy uses (e.g. nightlights, crops)
- Energy resource availability (e.g. GHI / solar potential)
- Power infrastructure (e.g. existing and planned grids and minigrids)
The NLP-powered chat assistant will provide tailored prompts and real-time support, ensuring users can optimize their use of EAE's features. By breaking down barriers of language and technical expertise, this innovation aims to enhance EAE's accessibility and usability, driving broader adoption and amplifying its impact in advancing inclusive energy transitions. It should serve as an always-available assistant that simplifies navigation for new users and enhances analytical outcomes for returning users.
A new repository being created and populated to:
- Integrate data and train the LLM.
- Annotate data for intents and responses to train the model effectively.
- Fine-tune a pre-trained LLM on a curated dataset (e.g. OpenAI, T5).
- Provide validation and accuracy measurements.
- Design an interactive chatbot and integrate it into EAE.
- GenAI backend (API key) — please use only open-source LLMs for this exercise.
- An API to connect the GenAI to the database or the frontend (depending on the chosen solution).
- The frontend of the chatbot.
EAE is written in plain/modern JavaScript (ECMAScript 2020). There is no framework; instead a traditional C-style programming pattern is enforced. The directories contain:
src: JavaScript codestylesheets: CSS codeviews: HTML documentsbin: scripts and executables
- Experience with Generative AI tools (RAG, Open Knowledge Format, prompt engineering, or chatbot integration).
- Skills in geospatial data analysis using PostGIS, QGIS, GeoJSON, or spatial SQL.
- Proficiency in backend development and API integration.
- Familiarity with web mapping and UI tools.
- Santiago Sinclair Lecaros
- Akansha Saklani
- Join the OSEAS Discord server: https://community.oseas.org/
- Introduce yourself in the
#introductionschannel and join the relevant channels for this challenge. - For physical participants: bring a computer (and required adapters) for some hacking.
- Read the documentation:
We will provide examples of user queries with expected output (as training datasets) to the registered participants.