OASIS

Inspiration

Responders often face heavy administrative burdens when documenting and reviewing incident reports. These reports increase in size as disaster unfolds. Responders have to scroll through lengthy documents while under time pressure making it difficult to get a complete picture of the situation leading to delayed responses and poor decision making

What it does

OASIS app lets users generate disaster incident reports by selecting incident type (e.g. flood, landslide, typhoon), location, and date. It provides instant summaries, resource deployment details, and recommendations using RAG powered by GPT-OSS, and includes an interactive chatbot for deeper exploration of the reports.

How we built it

We built OASIS using Streamlit UI and Python-based backend to ensure full offline capability and to provide a seamless and user-friendly experience. For AI functionality, we integrated GPT-OSS-20B model capable of running on laptop hardware, paired with a local vector database for incident reports for retrieval of the latest available information. We followed a 3-week sprint plan focused on prototyping, demo impact, and resilience-first design.

Challenges we ran into

Designing for a fully offline application meant rethinking assumptions around connectivity, storage, and AI inference. We had to optimize prompt content and data pipeline to ensure the LLM could run locally without draining too much resources. Of course, building something that feels intuitive for responders under stress was a constant design priority for UI features.

Accomplishments that we're proud of

In under a month, we built a working MVP that generates consolidated reports from available incident data. Through real disaster scenarios, we demonstrated how OASIS can cut report consolidation time from hours to minutes. We also developed a user-friendly UI to support users. Most importantly, we created a tool designed to help communities recover more effectively during their most vulnerable times.

What we learned

We realized how AI can be the most powerful not when it’s flashy, but when it quietly fills gaps in human workflows. We also deepened our understanding of disaster protocols, local governance, and the real-world needs of disaster responders. Most importantly, we learned that even in a hackathon, building with humanity in mind makes every technical decision sharper and more meaningful.

What's next for oasis-gpt-oss

Our team envisions a more comprehensive offline-first app that goes beyond drafting reports. The envisioned offline-first system would allow responders to log incidents through structured and free-text inputs, attach photos or videos, and capture GPS coordinates, even without internet access. Data would be stored locally and later synced with LGU or national disaster systems once connectivity is restored. To better capture and process incident data, we plan to add multi-language support so that reports and logged incidents in local languages can be understood and generated effectively. We also aim to fine-tune oasis-gpt-oss on disaster-specific datasets to further improve the accuracy, relevance, and consistency of generated reports.