trend-to-supply

Inspiration## Inspiration

This started from a familiar ecommerce problem: ideas are cheap, but validation is slow. A founder can have a rough idea like “home fitness” or “desk setup,” then spend days jumping between TikTok, Reddit, Amazon, Google, supplier marketplaces, spreadsheets, AI chats, and email.

We wanted to build the tool we wished existed: something that could take a vague product category and turn it into a real, evidence-backed sourcing opportunity.

What it does

Trend-to-Supply is an agent that turns a market vertical into a sourced product bet.

You enter a category, and the agent finds trend signals, reads customer pain points, validates demand, discovers suppliers, scores the opportunity, designs a better product variant, generates concept images, and drafts supplier RFQ emails.

It is meant to feel like a small product research team working in one workflow.

How we built it

We built the app with Next.js and a live dashboard so users can watch the agent work step by step.

Google ADK orchestrates the agent pipeline. Apify collects signals from TikTok, Amazon, and Reddit. Tavily validates demand and searches supplier marketplaces like Alibaba, Made-in-China, Thomasnet, and Global Sources. Gemini turns messy data into structured insights, supplier candidates, scores, and RFQ drafts.

We used Nano Banana / Gemini Image for product concept visuals, ClickHouse to store the full audit trail, Prometheux for symbolic stockCandidate reasoning, and Gmail to create human-approved supplier drafts.

Challenges we ran into

The hardest part was making a real multi-tool agent reliable. Live data is messy, supplier pages are inconsistent, APIs can fail, and AI outputs need strict validation.

We also wanted the agent to feel trustworthy, not like a black box. That meant streaming every step into the UI and saving the same events into ClickHouse so the process could be inspected later.

Accomplishments that we're proud of

We are proud that the agent does more than generate ideas. It produces a full sourcing workflow: evidence, pain points, supplier candidates, scores, product visuals, and outreach drafts.

We are also proud of the transparency. You can see each tool being used in real time, from Apify scrapes to Tavily searches, Prometheux reasoning, ClickHouse writes, and Gmail draft creation.

What we learned

We learned that useful agents need more than a good prompt. They need tools, memory, schemas, retries, reasoning, and human checkpoints.

We also learned that generative AI and symbolic reasoning work well together. Gemini is strong at understanding messy market data, while Prometheux helps turn facts into a more explainable verdict.

What's next for trend-to-supply

Next, we want to add deeper review analysis, richer supplier comparison, sample-cost estimation, margin forecasting, and follow-up tracking after Gmail drafts are created.

Long term, Trend-to-Supply could become an operating system for ecommerce founders: from idea discovery to supplier outreach to launch planning.

Built With

• adk
• claude
• clickhouse
• cursor
• gemini
• google
• nano
• prometheux
• python
• react
• tavily
• typescript