Inspiration
What it does## Inspiration
Meetings contain important decisions, responsibilities, risks, and follow-up actions, but turning long recordings into reliable professional documents is still slow and labor-intensive. Generic transcription tools produce text, yet rarely understand the business scenario or generate reports that match an organization’s required format.
We created ZhuTing AI Meeting Platform to transform spoken conversations into structured, actionable, and professionally formatted knowledge.
What it does
ZhuTing provides an end-to-end intelligent workflow:
- Captures live microphone audio or imports existing recordings
- Automatically transcribes speech in real time
- Identifies the most appropriate business scenario
- Generates structured summaries, decisions, risks, and action items
- Produces scenario-specific Word and PDF reports
- Supports meeting minutes, management briefings, personal memos, interviews, and customer visits
- Enables natural-language Q&A over current and historical meetings
- Preserves user profiles, meeting history, and generated deliverables
Instead of merely converting speech into text, ZhuTing converts conversations into usable organizational assets.
How we built it
The platform uses a modern web architecture with a React and TypeScript frontend, a FastAPI backend, and SQLite-based persistence. It integrates Alibaba Cloud speech-recognition services for transcription and Qwen models for scenario classification, semantic analysis, report generation, and meeting Q&A.
Our processing pipeline is:
- Audio acquisition or file upload
- Speech recognition and transcript normalization
- Multi-signal scenario classification
- Structured information extraction
- Scenario-specific report generation
- Word and PDF rendering
- Historical storage and contextual retrieval
The application is containerized with Docker and deployed on Alibaba Cloud behind Nginx. Responsive layouts support desktop and mobile browsers.
Challenges we ran into
One major challenge was distinguishing between scenarios that share similar vocabulary. A customer visit may discuss projects and schedules, while an engineering meeting may also mention organizations and cooperation. Keyword matching alone was not reliable.
We therefore designed a hybrid classification approach that combines domain evidence, conversation structure, semantic model judgment, conflict resolution, and calibrated confidence scores.
Another challenge was generating documents that preserve strict corporate templates. We developed a template-aware rendering pipeline that maps extracted content into predefined sections, tables, headers, logos, and pagination rules while keeping Word and PDF outputs visually consistent.
Cross-platform microphone access was also challenging because modern browsers require HTTPS and explicit user permission. We designed the recording experience around secure browser APIs and clear device states.
Accomplishments that we're proud of
- Built a complete audio-to-report workflow
- Created specialized reporting templates for multiple business scenarios
- Implemented automatic scenario detection with explainable confidence
- Added contextual Q&A across historical meetings
- Delivered editable Word and layout-consistent PDF exports
- Created a responsive interface for desktop and mobile use
- Deployed the complete platform through a production container stack
What we learned
We learned that high-quality meeting intelligence depends on more than a powerful language model. Reliable results require strong scenario definitions, structured extraction rules, confidence calibration, document engineering, human review paths, and careful interaction design.
We also learned that professional users value traceability. Generated conclusions should remain connected to the original transcript, and uncertain information must be clearly marked instead of invented.
What's next for ZhuTing
Next, we plan to improve speaker diarization, multilingual transcription, domain terminology adaptation, semantic search across larger meeting collections, collaborative review, task-system integration, and organization-level knowledge analytics.
Our long-term goal is to make every important conversation immediately searchable, actionable, and ready for professional delivery.
How we built it
Challenges we ran into
Accomplishments that we're proud of
What we learned
What's next for ZhuTing AI Meeting Platform
Built With
- ai
- alibaba
- api
- cloud
- dashscope
- docker
- fastapi
- generative
- language
- libreoffice
- natural
- nginx
- processing
- python
- qwen
- react
- recognition
- responsive
- rest
- speech
- sqlite
- typescript
- web
- websocket
Log in or sign up for Devpost to join the conversation.