Wildfire AIM (AI Mitigation)

We used AI and Microsoft Azure to build a RAG (Retrieval Augmented Generation) Search Chat Application. Its focus is on promoting prescribed burning as a wildfire prevention and mitigation method, as practiced by the Karuk Tribe of Northern California. Our system provides culturally sensitive guidelines for US government agencies to expand the use of prescribed burning in cooperation with native tribes.

#AI #microsoftazure #wildfireprevention #prescribedburning #hackathon

Project repo: https://github.com/ThanaReka/WildfireAIM

Transcript

1. Wildfire AIM Wildfire AI Mitigation: Using AI to help mitigate and prevent wildfires Microsoft AI Innovation Challenge June 2025 Team: Vui Nguyen, Reka Thanabalan
2. Overview • Introducing the Team • The Challenge We’re Solving • The Problem Statement • The Solution • Key Learnings • Future Development
3. The Team
4. The Team Vui Nguyen: AI Solutions Architect Reka Thanabalan: AI Engineer
5. The Challenge
6. The Challenge Wildfire Resilience: • Help emergency responders, planners, and policymakers mitigate the dangers of wildfires
7. Problem Statement
8. Problem Statement • Too much data on wildfire management • Those in charge of implementing wildfire prevention and mitigation: ◦ Fire Agencies, Bureau of Land Management, etc ◦ How to sift through all that data and develop an actionable plan? • How can AI be used to solve this problem?
9. The Solution
10. The Solution • RAG (Retrieval Augmented Generation) Search Chat Application • Focus on: ◦ Prescribed burning as a wildfire prevention and mitigation method ◦ As practiced by the Karuk Tribe of Northern California ◦ Provides culturally sensitive guidelines for US government agencies to expand the use of prescribed burning in cooperation with native tribes • Possible users of application: ◦ Bureau of Land Management ◦ Fire Agencies ◦ Forest Service ◦ Other US government agencies
11. Project Demo Video link: https://bit.ly/WildfireAIM-Demo
12. Web Front End - Screenshots
13. Web Front End - Screenshots
14. Architecture
15. Azure Services Used • Azure OpenAI • Azure AI Search • Azure Storage
16. Tech Stack • Frontend: Python, Bootstrap • Backend: Python, Flask • Azure AI Backend: Configured through Azure Portal
17. Challenges and Workarounds • Challenge: Hitting API rate limits when calling OpenAI endpoint to generate response • Workarounds / solutions that we tried: ◦ Tried to get load balancing working ▪ This didn’t work and ran out of time ◦ Request an increase in quota so we don’t hit rate limits ▪ Had to request increase in quota in advance and ran out of time
18. Challenges and Workarounds • Challenge: Hitting API rate limits when calling OpenAI endpoint to generate response • Workarounds / solutions that we tried: ◦ Got answers from index chat playground alone ▪ Not getting “full” quality answer from index + OpenAI chat ◦ Got answers from the AI Foundry chat playground alone ▪ Not getting “full” quality answer from index + OpenAI chat
19. Challenges and Workarounds • Final Workaround That We Chose: ◦ Chose the index search answers ◦ Hardcode answers from index chat playground into the web front end • Best Solution out of Workarounds Because: ◦ While we are missing the addition of OpenAI chat in the generated response, ◦ It was important for our responses to be grounded in our data for quality control and improving the chat narrative
20. Key Learnings
21. Key Learnings • Finding balance between spreading benefits of prescribed burning as wildfire prevention with cultural sensitivity (avoiding cultural appropriation) • For proof of concept, narrowing scope is important • Chose to focus on one method of wildfire prevention (prescribed burning) and dive deeper into its cultural importance to the Karuk Tribe • With time running out on project, need to focus on: ◦ What we can get done ◦ Implement workarounds as needed ◦ Put the rest into the future roadmap
22. Responsible AI • Since this is a new project, we have only started on our Responsible AI Journey • Here is how Wildfire AIM is implementing the Microsoft Responsible AI principles • Fairness and Inclusiveness: ◦ By focusing on recommending the use of prescribed burning in culturally sensitive ways, and avoiding cultural appropriation • Transparency: ◦ Data sources used to ground the RAG responses are documented in our Github repo • Privacy and security: ◦ API keys are stored securely and separately from public code ◦ System currently does not handle sensitive user data • Reliability and Safety, and Accountability: ◦ Continued testing during development will ensure continued quality of our system
23. Future Development
24. Future Development • Find way to fix API rate limiting problem so the application can fully work end to end without workarounds • Convert one-shot search to multi-turn chat
25. Thank you! Questions? • Github repo: https://github.com/ThanaReka/WildfireAIM • Team: Vui Nguyen, https://github.com/vuinguyen • Team: Reka Thanabalan, https://github.com/ThanaReka