Inspiration
the inspriation came from building production aware systems as senior year student to get expsoure into agentic ai and backend operations and building a secure system around it
What it does
SENTINEL is an autonomous research agent that investigates a topic end-to-end — searching live web sources, deepening findings across multiple references, identifying contradictions between sources, and producing a verified written report without human input after the initial prompt.
How we built it
built it using Python, Gemini 2.0 Flash, Tavily search API, SQLite, and Sentence Transformers.
Challenges we ran into
the learning curve for this project was very steep and found myself struggling to make a proper mental model was to be formed , but later thorugh dedication and trial and error was able to overcome , found troubles in implementing the backend logic and lot of fall backs calls and execption handling , since it was a first time implmenetation
Accomplishments that we're proud of
The core contribution is the trust system built around the agent. Every action is recorded in a SHA-256 tamper-evident event chain. Tools are phase-gated — the agent cannot call tools outside its permitted phase, enforced in the dispatcher not the prompt. Tool outputs are sanitized for prompt injection before reaching the LLM. The plan is hashed at creation and verified before every phase transition. A loop detector prevents infinite repetition and the system always exits in a defined state — success, inconclusive, or halted.
What we learned
how a true production aware system is used and what the real world expects of us in as software developers
What's next for autonomus research agent with trust system
making them more robust with scalablity , advanced api design , json-> toon conversions , latency are up our alley for this project
Built With
- and
- built-with-python
- gemini-2.0-flash
- sentence
- sqlite
- tavily-search-api
Log in or sign up for Devpost to join the conversation.