← Back to projects

Trend Pilot

An agentic AI assistant that helps small creator teams catch trends early, decide what to film, and ship as a team - the multi-agent "Agentic Trend Orchestrator," owned end-to-end from architecture through CI/CD.

The problem

The creator economy is a $1.3 trillion opportunity, but the tooling targets big brands and agencies. Roughly 60% of creators are small teams who get left behind. By the time a 2-3 person team spots a trend, debates it over Slack, and coordinates a shoot, the idea has changed and the trend is gone - with messy workflows and unclear ownership along the way.

Who it's for

What I built

How it works

A multimodal trend-detection pipeline ingests audio, video, and text, transcribes and embeds it, and stores vectors in Milvus-lite. Retrieval feeds a GPT-4o VLM reasoning layer that drives recommendations, scripts, and task orchestration. The product is a multi-agent system with persistent memory built on FastAPI + Next.js + Docker and shipped through a CI/CD pipeline.

The value (illustrative)

A modeled scenario from the product pitch - a food creator with a team of three:

1.2M views
in 26 hours after a trend was caught - vs 8K views in 4 days the month before
~22x ROI
modeled: +$3.3K/mo and ~60 hours saved
25% → 60%
hit rate on posts
10 → 5 hrs
time per post

Tech

FastAPINext.jsDockerPythonTypeScriptGPT-4o VLMMilvusRAGMulti-AgentCI/CD

Full project report

Product deck covering the problem, target users, multi-agent architecture, GPT-4o VLM trend pipeline, and modeled ROI for creator teams.

Download PDF