Silent Meeting Agent

• 

  Logo

• 

  Interface of the tool

• 

  Outputs of the upload video generated by our tool

Inspiration

Meetings take up a huge amount of time, and most people don’t actually need the full recording — they only need the moments where their name was mentioned, tasks were assigned, or important decisions were made. We were inspired by a simple question: “What if an AI could attend the entire meeting for you and deliver only the parts that matter?” That idea became Silent Meeting Agent — a tool that transforms long meeting recordings into clear, actionable outcomes.

What it does Silent Meeting Agent analyzes a meeting video/audio and generates a short, professional report containing: 30-second highlight recap (most important moments) Mentions of the user’s name with timestamps Action items (task + owner + deadline) Key decisions made Risks, blockers, and unanswered questions A ready-to-send follow-up message for the user

How we built it We built the project using Gemini in Google AI Studio as a multimodal meeting intelligence agent,

Key steps: User uploads a meeting recording (video/audio) User enters their name (e.g., Prashuuu) Gemini analyzes the content and generates a structured output We optimized the prompt to prioritize:

timestamps accountability clear tasks over long summaries Challenges we ran into The AI initially produced long summaries instead of actionable outputs → Solved by enforcing a strict format and limiting each section.

Meetings often have unclear task assignments → Added “Owner / Deadline / Confidence” fields to improve reliability. Missing details like deadlines or responsible people → Instead of guessing, the agent outputs “Not mentioned” to avoid hallucinations. Making the demo judge-friendly → Focused on the “wow factor”: name mentions + tasks + timestamps. Accomplishments that we're proud of Built a real meeting productivity agent, not just another chatbot Created an output that feels like a manager-ready action report Made it demo-ready within seconds: upload → analyze → results Designed a clean project identity: name + logo + tagline

What we learned Multimodal AI is most powerful when it produces decisions and actions, not just summaries Prompt structure directly impacts output clarity and quality Real meetings are messy, so accuracy and reliability matter most Trust improves when the model clearly says “Not mentioned” instead of guessing What's next for Silent Meeting Agent Next upgrades we plan to add: Auto-create Jira tickets from action items Calendar event generation for deadlines and follow-ups Slack / WhatsApp sharing for instant updates A “Only my name moments” highlight reel mode Support for meeting chat, slides, and documents as extra context

Built With

• css
• gemini
• html
• javascript