Spark AI Platform
Enterprise generative AI hub for SAE International
Spark AI is the internal generative AI hub built for SAE International. One login gives teams 46 AI models across six providers, 40+ specialized tools, and a builder for custom assistants. It runs in production on Supabase, Netlify Functions, and Railway agent runtimes.
The Challenge
SAE International staff were reaching for scattered consumer AI tools. No shared access, no governance, no institutional memory. The organization needed one place where every team could use AI safely.
Each AI provider means separate keys, billing, and interfaces — unmanageable for non-technical staff
Blank chat boxes don't fit specialized work like standards research, HR writing, or document analysis
New models ship monthly; hardcoded model IDs go stale and break tools silently
Departments needed purpose-built AI tools without each one becoming a separate IT project
AI adoption needs teaching and guardrails, not just API access
The Solution
One hub behind one login. Direct chat with Anthropic, OpenAI, and Gemini models, 40+ specialized tool cards, a drag-and-drop assistant builder, and agentic services that take on real departmental work.
Multi-provider model hub: 46 model IDs across 6 providers, tracked in a single registry audited weekly by an automated agent
AI Assistant Builder: drag-and-drop composition of custom assistants with prompts, guardrails, memory, and voice — users build their own
Document Chat: upload documents, chat across them, and generate audio podcast summaries
Realtime Standards Consultation: upload a technical standard and the AI navigates, highlights, and explains it live with voice and vision
Arc Agent: a persistent AI agent with memory, tools, and voice on its own Railway runtime
Request a Build: employees submit work requests into an agentic build pipeline, with a human approving every plan
The Results
Spark AI runs in production for SAE International teams. It grew from three chat cards into a governed platform of 40+ tools, custom assistants, and agentic services — on lean serverless infrastructure.
Production Platform
Live and serving SAE International teams, with active development through June 2026
Governed Model Registry
46 model IDs across 6 providers in one source-of-truth file, audited weekly by an automated agent that catches stale models
From Chat to Agents
Grew from simple chat access into specialized tools, custom assistants, and long-running agent runtimes
Employee-Driven Intake
Request a Build routes any employee's request into an agentic factory pipeline with human approval gates
Gallery
Technical Architecture
React and Vite single-page app with Supabase for auth and data. Every AI call routes through Netlify serverless functions, so provider keys never reach the browser. Long-running agentic work moves to dedicated Railway runtimes built on the Claude Agent SDK. A single registry file governs all model IDs and is audited weekly by an automated agent.
Technology Stack
