Echo: Turning Every News Article and Grant Announcement Into Structured Startup Ecosystem Data, Automatically

Brief

Echo’s platform helps foundations and boards track grants, applications and impact across the WA startup ecosystem, and that means keeping tabs on every founding, funding round, partnership and award as it happens. We built an AI agent that reads any URL, sitemap or block of text, whether it’s a news article, a grant recipients list or an awards announcement, and turns it into structured event data ready to drop into Echo’s database. It identifies who was involved, what happened, when and where, matches every person and organisation against Echo’s existing records, and creates new entries when it finds someone new. Delivered as a set of private npm packages that slot straight into Echo’s existing codebase.

I have used Sam and Sean’s services multiple times for AI related software integrations. I have found them to be extremely capable, great communicators and deliver exceptional results. I would highly recommend them!

Nate Sturcke, Spacecubed

Problem

The WA startup ecosystem moves constantly: companies get founded, grants get awarded, accelerators run cohorts, awards get handed out. Every one of those moments matters to the foundations and boards using Echo, but each one is buried in a different news article, government press release or awards page, written in plain English with no structure to it at all.

Before this, keeping Echo’s records current meant someone manually reading through these sources, working out who was involved, checking whether that person or company already existed in the database, and typing it all in by hand. Realistically, that either ate hours every week or it simply didn’t happen, and the platform’s picture of the ecosystem quietly went stale. Echo needed something that could read any source thrown at it and do that matching and logging automatically, reliably enough to trust without a human checking every line.

Solution

Feed the system a URL, a sitemap, a list of URLs, or plain text, and it reads the content, works out what kind of event happened (a company founded, a grant awarded, an award won, and 17 other categories), and pulls out who was involved, when, and where. It then checks those people and organisations against everything already in Echo’s database, matching what exists and creating new entries for what doesn’t, before writing the finished event straight into the platform.

• Built by an AI-native team, not just developers: this wasn’t a standard integration job. It’s a genuinely AI-heavy build, and even a founder who codes himself needed a team that lives and breathes how these models actually behave to get it right.
• Matches, not just extracts: most extraction tools stop at pulling out names. This one checks every entity against Echo’s existing database and only creates a new record when there’s genuinely no match, keeping the data clean instead of duplicated.
• Handles messy, real-world sources: sitemaps, single URLs, arrays of URLs, or raw text all work through the same pipeline, built to cope with the wildly inconsistent way news and grant announcements are actually written.
• Modular by design: event extraction and fuzzy matching were shipped as separate npm packages, so Echo can use, update or extend either piece independently.
• Tested and refined against real sources: every version was benchmarked against real news articles and grant pages before being signed off, not just clean test data.

Result

Across rounds of testing on real sources, the fuzzy-matching engine came back at 97 to 100 percent accuracy, and on clear, well-structured sources like grant recipient announcements, event extraction hit a 100 percent success rate. Where sources got genuinely messy or ambiguous, accuracy dropped, exactly the kind of edge case a human would struggle with too. Work most foundations would have quietly stopped doing now happens without anyone lifting a finger, keeping Echo’s picture of the WA startup ecosystem accurate as it happens rather than a snapshot of whoever had time to update it last.

Under the Hood