AI Chat Platform

Overview

This platform provides Gosign's internal teams with a unified chat interface for interacting with multiple large language models. Built as a self-hosted solution, it gives the organization full control over their AI tooling — no data leaves company infrastructure.

Problem

The team was using multiple disconnected AI tools — different interfaces for different models, no shared conversation history, and no way to compare model outputs side by side. Each tool had its own authentication, its own data storage, and its own limitations. There was no central place to manage prompts, documents, or team knowledge.

Solution

We built a unified chat platform that abstracts away the differences between LLM providers behind a single, polished interface. Users select a model, start a conversation, and the platform handles routing, authentication, and response streaming transparently. Document upload and retrieval-augmented generation (RAG) capabilities let users ground conversations in company-specific knowledge.

My Role

I was responsible for the frontend interface and Supabase integration. I built and customized the chat UI components, implemented real-time message streaming, designed the document upload and management interface, and integrated Supabase for user authentication and data persistence. I also contributed to the Docker-based deployment pipeline.

Architecture

• Frontend (SvelteKit): Reactive chat interface with real-time streaming, file uploads, and model selection
• Backend (FastAPI / Python): Handles LLM routing, RAG pipeline, and document processing
• Supabase: User authentication, conversation storage, and file management
• Docker: Containerized deployment for consistent environments across development and production
• Multi-model support: Abstracts OpenAI, Anthropic, and local model APIs behind a unified interface

Impact

The platform became the default AI tool for the entire Gosign team. Centralizing model access eliminated fragmented tool subscriptions and gave the organization visibility into how AI was being used. The RAG pipeline allowed teams to query internal documents directly in chat, reducing time spent searching for information.