AI-Powered Real-Time Collaborative Workspace

A distributed, microservices-based platform enabling real-time team collaboration with an integrated, context-aware AI Assistant. This system utilizes a Retrieval-Augmented Generation (RAG) pipeline to ground AI responses in private project data and conversation history.

---

System Architecture

The platform is built on a Polyglot Microservices Architecture, separating real-time traffic handling from heavy AI inference tasks.

Core Services

• User Service (Go)

  • Manages identity
  • JWT-based authentication
  • Persistent group (room) membership

• Collaboration Service (Go)

  • Handles high-concurrency WebSocket connections using a custom Hub pattern
  • Manages room namespacing
  • Persists chat history to PostgreSQL

• LLM Orchestrator (Go)

  • Acts as the system "brain"
  • Coordinates RAG retrieval
  • Assembles augmented prompts
  • Manages asynchronous events via Redis

• LLM Inference Service (Python)

  • Hosts a gRPC server
  • Performs semantic embedding (Sentence-Transformers)
  • Streams responses from Google Gemini 1.5 Flash

Infrastructure Layer

• Nginx – API Gateway and WebSocket reverse proxy
• Redis – Low-latency Pub/Sub for real-time AI token streaming
• Apache Kafka – Durable event log for background AI processing
• PostgreSQL + pgvector – Relational + vector database for hybrid storage

---

Key Features

1. Intelligent Collaboration

• Real-Time Messaging

  • Bi-directional communication with near zero latency

• Namespaced Rooms

  • Multi-tenant isolation (e.g., #science, #dev)

• Persistent Groups

  • Password-protected rooms
  • Join once → access forever

---

2. Context-Aware AI (RAG)

• Hybrid Context Awareness

  • Chat history (PostgreSQL)
  • Documents (pgvector embeddings)

• Streaming Responses

  • Token-by-token delivery via WebSockets
  • Real-time typing effect

• Passive AI Observation

  • Kafka-based background worker
  • Generates automatic Room Snapshots (TL;DRs) every 10 messages

---

3. Dynamic Knowledge Management

• Live Training

  • Users can teach the AI directly from UI

• Document Ingestion

  • Supports PDF & TXT uploads

• Smart Chunking with Overlap

  • Prevents context loss during embedding

---

4. Personal Copilot Sidebar

• Private AI chat interface
• Ask questions about room context
• Keeps main chat clean and focused

---

Tech Stack

Backend

• Go (Golang)
• Python

Frontend

• React
• Vite
• Tailwind CSS v4
• Lucide Icons

Communication

• gRPC
• Protobuf
• WebSockets
• Redis Pub/Sub
• Apache Kafka

Database

• PostgreSQL (pgvector)
• Redis (Caching)

Deployment

• Docker
• Docker Compose
• Nginx

---

Getting Started

Prerequisites

• Docker & Docker Compose
• Google Gemini API Key

---

Installation

1. Clone the Repository

git clone https://github.com/yourusername/ai-collab-system.git
cd ai-collab-system

2. Environment Setup

Create a .env file in the root directory:

GEMINI_API_KEY=your_actual_key_here
DB_DSN=postgres://admin:secretpassword@postgres:5432/collab_db?sslmode=disable
REDIS_ADDR=redis:6379
KAFKA_ADDR=kafka:9092

3. Launch the Stack

docker-compose up -d --build

4. Access the Platform

• Frontend (via Nginx):
  http://localhost:8080

• Or your Vite dev server port

---

Core Engineering Lessons

• Cross-Language Integration

  • Bridging Go & Python via gRPC
  • Combines performance (Go) with ML ecosystem (Python)

• Event-Driven Design

  • Redis & Kafka for decoupling services
  • Keeps UI responsive during heavy AI tasks

• Vector Search

  • Semantic retrieval using pgvector
  • Solves the "small context window" limitation of LLMs

---

Future Improvements

• Role-based access control (RBAC)
• Multi-modal document ingestion (images, audio)
• Fine-tuned local LLM support
• Advanced analytics dashboard