Meet2Code

meet2code

Inspiration

Developers spend a lot of time in meetings discussing features, bugs, and architecture decisions. But after the meeting ends, someone will still have to manually translate those conversations into GitHub issues, implementation plans, and finally code.

Some ideas may get lost in meeting notes, and valuable developer time is wasted on documentation rather than on the building.

So we asked: What if we can turn meetings could automatically turn into code workflows?

What it does

Meet2Code listens to meeting audio in real time and automatically converts discussions into structured development workflows.

It can:

Transcribe live meeting audio
Detect decisions, tasks, and feature discussions
Generate structured development tasks
Create GitHub issues, code, and pull requests.
Help teams move directly from discussion to development

Instead of manually writing tickets after meetings, Meet2Code turns conversations into actionable work instantly.

How we built it

Meet2Code is built as a real-time pipeline that converts live meeting audio into structured developer tasks and workflows.

The system is composed of three main layers: audio ingestion, AI task extraction, and developer workflow generation.

Audio & Transcription Pipeline

We capture local microphone audio and meeting audio streams using ffmpeg, with VBCABLE acting as a virtual audio device on system audio (Teams, Meet, Zoom, etc.).

A Node.js transcription service streams the audio to the AssemblyAI Streaming API (u3-rt-pro), which performs real-time speech-to-text and speaker diarization. This produces a continuous transcript of the meeting as it happens.

AI Task Extraction

Once transcripts are produced, a Python pipeline processes them to extract the actionable development tasks.

We use the OpenAI API (gpt-4.1-nano) to analyze transcript segments and convert them into structured engineering outputs such as:

feature requests
bug reports
implementation tasks
architecture decisions

These tasks are formatted into structured developer artifacts like Markdown task lists and JSON pipeline state files.

Developer Workflow Automation

To connect meeting insights directly to engineering workflows, we integrated developer tooling automation.

Using Anthropic Claude Code CLI, the system can autonomously generate code changes and prepare pull requests based on extracted tasks. We also use the GitHub CLI (gh) to manage issue creation, labeling, and PR workflows.

This creates a pipeline where meeting discussions can directly generate engineering work items.

Frontend Dashboard & Demo App

We built a monitoring dashboard using Next.js and React. The dashboard provides real-time visibility into the transcription and task extraction pipeline.

Challenges we ran into

One of the biggest challenges was capturing and streaming live meeting audio into our pipeline

We originally wanted to create a automated bot that joins meetings on various meeting platforms, but found that it would be unfeasible for the scope of this project as these platforms often try hardest to prevent bots. We then looked at browser-based audio capture by a meeting attendee but this introduced issues like muted playback and capturing turning off with tab movement. We instead built a Node.js-based local audio pipeline using FFmpeg to capture and stream audio more reliably.

Another challenge was turning messy, natural conversation into structured developer tasks. Meetings often include partial ideas, interruptions, and vague language. Designing prompts that extract clear actionable items required multiple iterations.

Real-time processing also required careful handling of streaming data and partial transcripts.

Accomplishments that we're proud of

In the scope of this hackathon we built a system that:

Captures live meeting audio
Transcribes it in real time
Extracts development tasks
Converts them into structured developer plans

What we learned

Real-time audio streaming systems
Speech-to-text pipelines
Prompt engineering for extracting structured information from messy conversations
Integrating Agentic AI solutions into real developer workflows

What's next for Meet2Code

Add a meetings tab to the dashboard that displays details of previously tracked meetings and their outcomes
Currently, our solution runs a local CLI-based AI, we would like to put this on a server for seamless access on any computer
Our Node.JS runtime is currently just command based, it could use a GUI.