AI Product Build

An AI agent that reads a website and writes the pitch.

OutreachOS is a personal product build: paste a prospect URL, and the tool reads their site, extracts context, and streams back a personalised cold email in seconds.

Client

Personal build, Gemini Live Agent Challenge

Engagement

Full-stack AI application

Stack

React, Gemini 2.5 Flash, Jina Reader

The problem.

Cold outreach dies on personalisation. Doing it properly means reading every prospect website, understanding their business, and writing something specific. That takes 15 to 20 minutes per lead, which makes real volume impossible for a solo operator.

The challenge: compress that entire research-and-write loop into an automated agent without producing generic AI slop.

What I built.

React frontend with clean single-input workflow: paste URL, get email
Jina Reader integration scraping and extracting prospect website content
Gemini 2.5 Flash generating personalised outreach grounded in the scraped context
Server-Sent Events streaming the output live as it generates
Submitted to the Gemini Live Agent Challenge on Devpost under the Creative Storyteller category

The stack.

React Gemini 2.5 Flash Jina Reader Server-Sent Events Devpost

Open source.

The full codebase is public. View the repository on GitHub →

The result.

A working end-to-end AI agent: data ingestion, context extraction, grounded generation, live streaming output. It demonstrates the exact architecture I build for clients who want AI working inside their own operations.