Inspiration
Early-stage investigations often involve fragmented information across scene notes, evidence logs, and witness statements. This is when important details can be missed or bias can be introduced. We built Anchor to provide a clear, neutral foundation—helping teams organize facts before drawing conclusions.
What it does
Anchor is an AI-powered investigation support tool designed to mirror how human investigation teams work—drawing on specialized analysis, aggregating information from multiple sources, identifying patterns, and learning from past investigative outcomes. It transforms unstructured inputs such as crime scene descriptions, evidence records, and witness testimonies into a structured starting report. Anchor highlights key facts, relationships, and gaps while preserving traceability to original sources, supporting investigators without replacing human judgment.
How we built it
Anchor uses a multi-agent architecture that mirrors real investigative teams. Text and image evidence—such as footprints, scene overviews, physical evidence descriptions, leads, and witness testimonies—are routed to specialized agents. A forensic RAG agent analyzes footprint images via similarity search to infer possible descriptors, a timeline agent reconstructs key events, a witness agent aggregates and analyzes testimonies, and a sketch agent generates suspect visuals from descriptions. All outputs are passed to a central detective agent that identifies patterns, compares findings against historical criminal data, and produces a structured, traceable investigative report.
Challenges we ran into
The main challenge was preventing over-interpretation. In investigative contexts, surfacing patterns without asserting conclusions is critical. Handling conflicting or ambiguous testimony while maintaining neutrality and traceability was also a key challenge.
Accomplishments that we're proud of
- Built a working prototype that meaningfully supports investigative workflows
- Designed outputs focused on clarity, neutrality, and trust
- Balanced technical capability with ethical responsibility
What we learned
We learned that AI is most effective in investigations when used as a structuring and support tool—not a decision-maker. Clear presentation and source transparency are essential for trust.
What's next for Anchor
Next steps include expanding Anchor with additional specialized forensic agents (e.g., bloodstain pattern analysis, cybercrime, ballistics), improving RAG-backed reference datasets for higher-quality evidence matching, and enhancing multimodal support. We also plan to add clearer timeline and entity visualizations, stronger auditability, and deeper integration with existing case management systems.
Log in or sign up for Devpost to join the conversation.