Skip to content

RudraBhungaliya/Ted

Repository files navigation

TED

Overview

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.


Objective

Enable developers and teams to receive relevant insights, summaries, explanations, and contextual assistance in real time without leaving their current workflow.


Core Capabilities

Real-Time Audio Processing

TED captures microphone and system audio streams, converts speech into text, and continuously processes conversations to maintain contextual awareness.


AI Developer Copilot

TED provides contextual assistance during development tasks, helping with:

  • Code explanations
  • API references
  • Debugging guidance
  • Architecture discussions
  • Documentation lookup
  • Technical problem solving

Meeting Intelligence

TED continuously understands ongoing conversations and provides:

  • Live contextual assistance
  • Action item extraction
  • Key discussion points
  • Automated summaries
  • Follow-up notes

Overlay-First Interface

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.


Streaming Response Pipeline

Responses are streamed token by token rather than waiting for complete generation. This significantly reduces perceived latency and improves usability during live interactions.


Context Structuring

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.


Caching Layer

Frequently repeated requests and session context are cached to reduce response time and API consumption while maintaining continuity.


History Tracking

Sessions, transcripts, and interactions are stored for later review, enabling searchable archives and post-meeting analysis.


System Architecture

Desktop Client (Electron + Next.js)

Responsible for:

  • Overlay interface
  • Audio capture
  • Screen capture
  • Session management
  • Streaming connections
  • Real-time interaction

Backend Server (Fastify)

Processes requests through a structured pipeline:

  • Speech-to-text
  • Transcript aggregation
  • Context construction
  • Prompt generation
  • LLM interaction
  • Streaming responses
  • Persistence and analytics

Shared Contracts

API schemas are shared across frontend and backend to maintain consistency and reduce integration errors.


Data Flow

Audio / Text Input

Speech-to-Text

Transcript Aggregation

Context Builder

Cache Lookup

Prompt Construction

LLM Processing (Streaming)

Streamed Response

Overlay Rendering

Session Storage


Design Principles

  • Low latency first
  • Overlay-based workflow
  • Context-aware assistance
  • Streaming responses
  • Persistent conversational memory
  • Minimal workflow interruption
  • Cross-application accessibility

Tech Stack

Frontend

  • Next.js
  • TypeScript
  • Electron
  • Web Audio APIs

Backend

  • Fastify
  • PostgreSQL
  • Prisma
  • Redis

AI

  • Gemini
  • Deepgram

Infrastructure

  • Streaming APIs
  • Session Persistence
  • Analytics Pipeline
  • Context Management

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages