Ooda Hack | Devpost

Inspiration

The OODA Loop: We wanted to speed up our decision making cycle from the bottom-up

What it does

Ingests multiple documents from OSINT sources Provide a synthesized summary of any given topic Provide an assessment against a threat assessment model Help speed up the daily SITREP

How we built it

Timed data pull from website using CRON job and web scraper (Beautiful soup) Processed documents with PDF reading python library PyMuPDF Extracted relevant data with Openai API / Cohere Summarized with Openai API (Got structured data using JSON mode) Aggregated and visualized in a streamlit app

Challenges we ran into

Finding good data sources for cyber reports Extracting data from the pdfs was a challenge due to context limits.

Accomplishments that we’re proud of

Discovery of a highly scalable problem with a compounding solution Extracting data on cyber threats in a detailed JSON format from text documents Learning MITRE framework

What we learned

The task: DISA’s problem scales far beyond The “threat assessment” model can be parameterized The team: we chose “team vibes” over idea We looked for highly complementary skill sets (AI/ML engineers, full stack, product/domain expertise) and wanted to make sure we passed “the airport test” We weren’t married to any particular problem we wanted to solve

What’s next for Ooda Hack

Automatically finding and scraping cyber attack sources from web Training an LLM on MITRE framework extraction from documents Scaling to other use cases with a scalable data ingestion, extraction and analysis API pipeline