TED is a real-time AI Meeting Assistant and Developer Copilot designed to provide immediate, context-aware assistance during meetings, discussions, coding sessions, and everyday development workflows.
Rather than acting as a traditional chatbot, TED operates as a low-latency overlay that stays alongside existing applications. It captures audio or text input, processes context through a structured backend pipeline, and streams responses incrementally to minimize interruptions and preserve focus.
The system is built around speed, contextual understanding, and seamless workflow integration.
Enable developers and teams to receive relevant insights, summaries, explanations, and contextual assistance in real time without leaving their current workflow.
TED captures microphone and system audio streams, converts speech into text, and continuously processes conversations to maintain contextual awareness.
TED provides contextual assistance during development tasks, helping with:
- Code explanations
- API references
- Debugging guidance
- Architecture discussions
- Documentation lookup
- Technical problem solving
TED continuously understands ongoing conversations and provides:
- Live contextual assistance
- Action item extraction
- Key discussion points
- Automated summaries
- Follow-up notes
TED runs as a lightweight overlay instead of replacing existing tools. Responses appear inside an always-on-top panel designed to stay out of the way while remaining instantly accessible.
Responses are streamed token by token rather than waiting for complete generation. This significantly reduces perceived latency and improves usability during live interactions.
Incoming data is first analyzed and structured before being sent to the model. Intent, domain, and conversation state are extracted to improve output relevance and consistency.
Frequently repeated requests and session context are cached to reduce response time and API consumption while maintaining continuity.
Sessions, transcripts, and interactions are stored for later review, enabling searchable archives and post-meeting analysis.
Responsible for:
- Overlay interface
- Audio capture
- Screen capture
- Session management
- Streaming connections
- Real-time interaction
Processes requests through a structured pipeline:
- Speech-to-text
- Transcript aggregation
- Context construction
- Prompt generation
- LLM interaction
- Streaming responses
- Persistence and analytics
API schemas are shared across frontend and backend to maintain consistency and reduce integration errors.
Audio / Text Input
↓
Speech-to-Text
↓
Transcript Aggregation
↓
Context Builder
↓
Cache Lookup
↓
Prompt Construction
↓
LLM Processing (Streaming)
↓
Streamed Response
↓
Overlay Rendering
↓
Session Storage
- Low latency first
- Overlay-based workflow
- Context-aware assistance
- Streaming responses
- Persistent conversational memory
- Minimal workflow interruption
- Cross-application accessibility
- Next.js
- TypeScript
- Electron
- Web Audio APIs
- Fastify
- PostgreSQL
- Prisma
- Redis
- Gemini
- Deepgram
- Streaming APIs
- Session Persistence
- Analytics Pipeline
- Context Management