BriefingLens

• 

  screenshot

• 

  screenshot

• 

  screenshot

• 

  mainpage

• 

  screenshot

Inspiration

Hackathon judges face a simple problem: too many videos, too little time. Watching every demo end-to-end is slow, and generic summaries often miss what matters for evaluation.

I built BriefingLens to turn raw demo footage into decision-ready, evidence-backed briefs — and then into a short multimedia “Briefing Story” reel that judges (and future audiences) can consume in seconds.

What BriefingLens does

BriefingLens is a video-first briefing workbench. Given a demo video, it:

• Extracts product positioning and judge notes (strengths, risks, uncertainties)
• Builds an evidence map: key claims tied to timestamps + keyframes
• Produces a judge-oriented heuristic scorecard (clearly labeled as draft scoring based on demo evidence)
• Generates a ≤60s “Briefing Story” highlight reel with optional AI narration (TTS)

It’s not a generic summarizer: the core is explainability and grounding — every important point is traceable back to the video.

Why this fits “Creative Storyteller”

BriefingLens uses Gemini to weave text, images, audio, and video into a cohesive mixed-media output:

• Text: structured judge brief (positioning, strengths, risks, open questions)
• Images: extracted keyframes as evidence cards
• Audio: optional narrated recap via Google Text-to-Speech
• Video: auto-rendered Briefing Story reel that interleaves narration + visuals + on-screen explanations

The result is a single, shareable storytelling artifact, not just a dashboard.

How We built it (tech)

Frontend

• Next.js / React

Backend

• FastAPI (Python)
• ffmpeg/ffprobe for video metadata + keyframe extraction
• Caching to avoid repeated Gemini calls on previously analyzed videos (save tokens)

AI / Cloud (Google)

• Gemini models via Google GenAI SDK
• Hosted on Google Cloud (Cloud Run deployment)
• Vertex AI / Google Cloud Text-to-Speech for narration (optional)

Key features (judge-first)

• Video-first: analyzes actual demo footage, not just slides/transcripts
• Judge-oriented: produces evaluation-ready insights and rubric-aligned notes
• Evidence-backed: timestamps + keyframes support every major claim
• Fast to review: compresses long demos into a brief + a short reel
• Transparent: flags uncertainties and open questions instead of over-claiming

Challenges & what we learned

• Timestamp trustworthiness: we added strict validation to ensure highlights never exceed video duration.
• Avoiding boilerplate: we removed static fallbacks and enforced video-specific Evidence / Interpretation / Open Questions grounded in the current job’s highlights.
• From “analysis” to “story”: turning structured evidence into a cohesive reel required a clean path from highlights → keyframes → script → render → narration.
• Demo UX: progress stages and caching were critical so reviewers don’t think the system is stuck and we don’t waste tokens on reruns.

Extensibility (beyond hackathons)

While this submission focuses on judges, the same workflow generalizes naturally:

• Sports: automatic highlight + commentary reels
• Esports / gaming: key moments + explainable scoring + narrated recaps
• Short-form video: auto-generated highlight compilations and story summaries

In short, BriefingLens is a reusable pipeline for video → evidence → story.

Built With

• fastapi
• ffmpeg
• gemini-(google-genai-sdk)
• google-cloud-run
• google-cloud-text-to-speech
• next.js
• react
• vertex-ai