Conference-Go

An AI-enabled smart conference connection matching and email generation system using Google Gemini API via AI Studio.

Features

• Preprocessing Module: Uses Gemini to parse natural language descriptions of people into structured JSON format optimized for embeddings.
• Embedding Module: Embeds structured person data into vector space using Gemini embedding models.
• Indexing Module: Brute-force indexing and similarity search for person embeddings.
• Attendee Profiles: 308 diverse fake person descriptions for testing and demonstration.
• Connection Matching: Input an ideal connection description and find the most similar people from the dataset.
• Email Drafting: AI-generated introduction emails between matched attendees.
• Chat: Converse with an AI persona of any attendee.
• Text-to-Speech: Listen to attendee responses via Gemini TTS.

Quick Start (Docker)

./setup.sh

The script checks for Docker, prompts for your API key (saves to .env), builds the image, and starts the server at http://localhost:8080. The SQLite database is mounted as a bind volume so profiles persist across container restarts.

Or manually:

export GOOGLE_API_KEY=your_key_here
docker compose up --build

Usage

Web Interface (Recommended)

Run the application in web server mode:

go run cmd/server/main.go -web

Or specify a custom port:

go run cmd/server/main.go -web -port 3000

Then open http://localhost:8080 in your browser.

The web interface provides:

• Browse all conference attendees
• Search for your ideal connection by natural language description
• Add new attendees via POST /api/attendees/add
• AI-generated introduction emails between matched attendees
• Chat with an AI persona of any attendee
• Text-to-speech playback of attendee responses

Command Line Interface

For command-line usage:

go run cmd/server/main.go "Your ideal connection description here"

If no description is provided, it uses a default example.

The system will:

1. Load and embed all 308 attendee profiles (database-cached after first run)
2. Process your ideal connection description
3. Find and display the top 5 most similar matches with details

Startup & Database

Profiles (structured data + embeddings) are stored in conference.db (SQLite) after the first run. On subsequent starts, only descriptions not already in the database are re-processed.

Scenario	Time (go run)
Cold start (empty DB)	~86s — preprocesses and embeds all 308 profiles with 10 concurrent Gemini API calls
Warm start (DB populated)	~24s — loads profiles from SQLite (most of this is Go compilation)

Once the database is populated, no API calls are made at startup. New attendees added via POST /api/attendees/add are persisted immediately and survive restarts.

Setup

1. Install Go 1.19+ if not already installed.

2. Get a Gemini API key:

   • Go to Google AI Studio
   • Sign in with your Google account
   • Create a new API key or use an existing one

3. Set environment variables:

   export GOOGLE_API_KEY=your-gemini-api-key

4. Install dependencies:

   go mod tidy

5. Run the application:

   go run cmd/server/main.go "Your ideal connection description here"

Architecture

• pkg/preprocessing/: Natural language → structured JSON conversion using Gemini.
• pkg/embedding/: Generates embeddings for person profiles using Gemini embedding API.
• pkg/indexing/: Brute-force vector indexing and cosine similarity search.
• pkg/database/: SQLite-backed profile store (structured data + embeddings).
• pkg/types/: Common data structures.
• pkg/email/: AI-generated introduction email drafting.
• pkg/conversation/: Attendee persona chat and text-to-speech via Gemini.
• pkg/data/: 308 fake attendee descriptions and evaluation cases.
• web/templates/: HTML templates for the web interface.
• web/static/: CSS and JavaScript files for styling and interactivity.
• cmd/server/: Main application supporting both CLI and web server modes.

.env support

Create a .env file in repository root:

GOOGLE_API_KEY=your_gemini_api_key

Install dependency:

go get github.com/joho/godotenv

Then run:

go run cmd/server/main.go -web

Your .env should be gitignored:

.env