Connecting scattered knowledge to answers with sources.

Useful business information is spread across documents, email, shared drives, and links. An AI interface becomes more useful when it can retrieve that information and show where its answer came from, rather than expecting someone to move context between tools manually.

Nucleus follows an ‘add once, access everywhere’ model. Its product brings documents, Gmail, Drive, and links into a shared knowledge layer that AI tools can query through MCP.

I worked across the Next.js product surface, the API layer, and the retrieval model. The project stack included TypeScript, PostgreSQL, Prisma, Node.js, Python, Tailwind CSS, and Shadcn UI.

My work was within an existing team and product. The scope described here is my full-stack contribution to the platform, rather than sole ownership of its ingestion connectors or every part of its infrastructure.

The interface, API, and retrieval model contribute to the same user task: asking a question and inspecting the supporting material. Source citations give users a way to follow an answer back to the underlying knowledge.

MCP access extends that knowledge beyond a single application interface. The same product idea—one source of truth, accessible from different AI tools—has to remain coherent across the web experience and the integration surface.

This engagement demonstrates full-stack contribution to an AI knowledge product, including the user interface, backend APIs, and retrieval layer. The public Nucleus website is linked for product context.

The description covers my 2024–2025 engagement. No retrieval accuracy benchmark, search latency, or customer productivity improvement is claimed; the current product may differ from the version I worked on.