Inspiration
Investigations today rely on fragmented evidence; video footage, text logs, and audio recordings that analysts must manually piece together. This process is slow, error-prone, and vulnerable to human bias. We wanted to explore how multimodal AI could assist investigators by reconstructing timelines and relationships automatically from scattered evidence.
The Urban Noir theme inspired us to design an investigative assistant that feels like a digital detective — helping uncover truth hidden across multiple sources of information.
What it does
Night Archivist is a multimodal AI investigative assistant that reconstructs events from fragmented evidence. Users upload video and text evidence through a dashboard, and the system extracts key events, identifies entities, builds a timeline, and generates a narrative case summary.
The platform visualizes relationships between actors and events while producing a narrated noir-style summary of the investigation.
How we built it
We built Night Archivist as a full-stack AI application.
The frontend dashboard was built using Next.js, allowing users to upload evidence and visualize investigation results through timelines and summaries.
A Python backend orchestrates the AI processing pipeline and integrates multiple AI services:
- TwelveLabs for video understanding
- Gemini for reasoning and narrative generation
- ElevenLabs for voice narration
The backend processes uploaded evidence, extracts structured events, constructs a timeline, and returns results to the frontend for visualization.
Challenges we ran into
Working with multimodal inputs required coordinating several AI services and ensuring outputs could be combined into a single timeline representation. Another challenge was designing a pipeline that could extract meaningful events from limited hackathon-scale evidence while still producing a compelling demo.
Balancing technical complexity with a reliable MVP within the hackathon timeframe was one of the biggest challenges.
Accomplishments that we're proud of
We are proud of building a complete end-to-end investigative workflow that processes multimodal evidence and transforms it into structured insights. Night Archivist successfully extracts events, constructs timelines, visualizes relationships, and generates narrative summaries in a single dashboard experience.
Integrating multiple AI systems into one cohesive investigative assistant during a hackathon was a major accomplishment.
What we learned
We learned how to orchestrate multiple AI APIs into a unified workflow and how multimodal AI systems can transform unstructured evidence into meaningful insights.
We also learned the importance of designing AI tools that support human reasoning rather than replacing it.
What's next for Night Archivist
Next, we want to expand Night Archivist into a more robust investigative platform. Future improvements could include support for additional evidence sources such as audio files, emails, and structured data, along with improved entity recognition and event extraction accuracy.
We also plan to enhance timeline visualization and enable collaborative investigation workflows.
Log in or sign up for Devpost to join the conversation.