Enterprise AI Assistant
Built a product-discovery AI assistant with streaming responses, inline side previews, and direct routing to product, company, and provider detail pages deployed inside an enterprise trade platform.
Overview
The platform had a large catalog of products, providers, sellers, and companies that users had to navigate manually. The solution was a conversational AI assistant built directly into the product, users could describe what they were looking for, and the assistant would surface the right product, its provider or seller, the responsible company, and all relevant data. Stats and charts were rendered inline, and a side preview panel let users explore entities without leaving the conversation. One click routed them to the full detail page when they needed to go deeper.
The Challenge
The assistant had to do more than answer questions, it had to act as a discovery layer for a complex, relational data model. Products have providers, providers belong to companies, companies have metrics. All of that had to surface coherently in a chat interface that felt fast and didn't overwhelm the user.
My Role
I owned the full frontend: chat architecture, streaming token renderer, tool-calling system, side preview panel, chart and stats rendering, routing integration, session state.
Architecture Decisions
Side preview panel with entity routing
When the AI surfaces a product, provider, or company, it populates a persistent side panel with a structured preview card. From there, users can navigate directly to the full entity detail page keeping the chat flow intact while enabling deep dives.
Tool-call driven rendering
The model triggered typed frontend components (product card, provider profile, company overview, chart block) rather than returning raw text — structured data stayed structured inside the conversation.
SSE streaming over REST
Used Server-Sent Events so tokens rendered as they arrived eliminating the frozen-loading feel and making the assistant feel responsive even on large responses with embedded data.
Impact
- check_circle
Users could find a product, its provider, seller, and company along with stats and charts through a single conversational query
- check_circle
The side preview panel eliminated page-switching for most discovery workflows, keeping users in flow
Tech Stack
Next.js · React · TypeScript · OpenAI API · Server-Sent Events · React Query · Axios · Tailwind CSS