SentinelAI-Beyond the Search

💡 Inspiration

We all know that AI can give answers in seconds. But how do we know whether those answers are actually trustworthy? During our brainstorming, we realised that finding information is only one part of an investigation. Checking whether the information is reliable, identifying contradictions, and understanding what is true are equally important.
 That idea inspired us to build SentinelAI — a system that doesn't just search for answers but also checks the evidence behind them.

🔍 What it does

->->🧠 Investigates, not just answers: Breaks complex questions into specialised AI tasks. ->->🔎 Verifies evidence: Checks sources and connects findings to supporting evidence. ->->⚖️ Detects contradictions: Highlights conflicting information and missing evidence. ->->🛡️ Built-in governance: Controls tool access and requires human approval for sensitive actions. ->->📊 Explainable reports: Presents findings, uncertainty, and audit trails in one place.

🛠️ How we built it

->->🤖 Multi-agent architecture: Separate agents handle research, verification, analysis, security, and reporting. ->->⚙️ Backend: FastAPI and Pydantic manage workflows and validate agent messages. ->->🎨 Frontend: React, TypeScript, and Vite power the investigation dashboard. ->->🔐 Policy Engine: Enforces permissions and records governance decisions. ->->🧪 Mock mode: Uses sample documents to demonstrate the investigation workflow.

.🚧 Challenges we ran into

->->🔗 Coordinating multiple agents while keeping their outputs consistent. ->->⚖️ Handling conflicting evidence without blindly trusting one source. ->->🛡️ Detecting malicious instructions hidden inside retrieved documents. ->->🔄 Connecting the frontend, backend, and investigation workflow. ->->📖 Making complex AI findings understandable and traceable.

.🏆 Accomplishments that we're proud of

->->💡 Built around a key question: Can we trust what AI tells us? ->->🤝 Designed specialised agents that work together throughout an investigation. ->->🔍 Combined evidence verification with contradiction detection. ->->🔐 Made security governance, human oversight, and accountability core design principles. ->->🚀 Took our first hackathon idea beyond a simple chatbot into an investigation platform.

.📚 What we learned

->->🎯 A confident AI answer is not necessarily a correct answer. ->->🔎 Evidence quality matters as much as the answer itself. ->->🤖 Each AI agent needs a clear role and validated outputs. ->->🛡️ Responsible AI requires security controls and auditability. ->->💻 Building SentinelAI strengthened our understanding of AI and full-stack development.

.🚀 What's next for SentinelAI — Beyond the Search

->->🌐 Integrate trusted real-world research sources. ->->🧾 Improve source reliability checks and evidence verification. ->->🛡️ Strengthen prompt-injection detection and security monitoring. ->->📈 Enhance live investigation tracking and interactive reports. ->->✨ Our vision: Don't just find answers. Discover what deserves your trust.

Built With

• artificial
• fastapi
• intelligence
• javascript
• multi-agent
• python
• react
• sqlite
• systems
• typescript