Portfolio Log #003

It's been an incredibly stressful week and it felt like not much noved. However, looking at my review notes, I can't believe I actually got so much done on the side. I have another post about managing time and attention in the works, but for now, let's check what I did this week.

Leadrack goes looking for its own customers

The biggest change this week was around email and identity resolution.

Leadrack can now do a better job of finding the person behind a lead and locating their email address. Email discovery is on by default for new agents, and there’s a button to try again manually when the first attempt comes up empty. I have another ace up my sleeve here that might find more and better leads again. But that's something for the next weeks.

Reddit results now link directly to both the post and the person who wrote it. This sounds small, but it makes those results much easier to understand and act on.

I also tested the Reddit agent myself, by asking it to find people who might need Leadrack. It found a ton and I reached out to five genuinely good leads via DM. That was probably my favourite moment of the week: using the thing for its actual job and getting results I would have been happy to receive as a customer.

There were also a bunch of smaller improvements to filtering, agent briefs and settings, including the long-overdue ability to delete a team.

Finding a sharper angle for Leadrack

I also found a direction that might help Leadrack stand apart from the growing pile of “AI GTM platforms.”

Most tools start with an ICP and try to find more people who match it. But the ICP itself is often still a guess.

The idea is to go one step earlier. A user describes who they think their best customer is, then Leadrack proposes three competing versions of that ICP and runs a pipeline for each one. Once there’s enough data, it can show which group actually responds best. More agents can then be assigned to the winner.

Instead of just helping find leads for an ICP, Leadrack could help figure out which ICP is worth pursuing in the first place.

It still needs to be tested, but it feels much more interesting than simply adding more agent types. I also think this would genuinely help me find the perfect ICP for the products by not locking in on an assumption early on.

Contentcron gets better at distribution

Contentcron now submits every successfully published article through IndexNow and connects to Bing Webmaster Tools, alongside the existing Google Search Console integration.

I also made some smaller improvements to how it writes changelog entries.

The social-post feature got a bigger rethink. Originally, the posts were tied directly to individual articles. That was a useful starting point, but it made them feel a bit like promotional leftovers.

Social posts are now their own thing. Contentcron learns the brand’s social voice and proposes a seven-day window of posts based on whatever is actually worth talking about, whether or not there’s an article attached.

That feels closer to the real goal: helping a product maintain a useful public voice, not just automatically posting “new blog post” twice a week.

Grepture gets smarter routing

Grepture got two new features this week.

Sessions now group related traces more clearly, making it easier to follow what happened across a sequence of model calls.

I also added rules for automatic model switching. Requests can now move to another model when credits are running low, when certain headers are present, or when another configured condition matches. This is pretty big (and cool) because it's a step to more routing capabilities and I could actually keep the whole established rule logic going.

There’s a new blog post too, partly because Grepture’s search traffic continues to point toward an interesting adjacent problem: PII redaction.

Scrubberduck might become a thing

I spent some more time exploring Scrubberduck, a possible standalone PII-redaction API spun out of Grepture. I've been toying with the idea for a while now, since all PII relatec content on Grepture by far outperformce other content.

I'm still not convinced it should be its own thing, but the SEO pull alone feels like I should try. I've been on the fence with this one for so long, I should probably just try releasing it.

The reality here, however, is also that getting very good at PII redaction is still really hard! Especially when supporting multiple languages.

I tested the current implementation against PII-Bench and I’m not convinced yet. There might be a useful product here, especially given Grepture's search traffic, but the benchmark results didn’t make the decision obvious. (It beat all in the paper and Presidio but it's far from 'we catch them all, always')

So it stays in the “worth tinkering with” pile for now.

The main theme this week was using the products instead of only building them. Leadrack found leads for Leadrack, Contentcron is being set up to market both products, and Grepture’s existing traffic is helping decide whether Scrubberduck deserves to exist.

Aiming for next week: run more Leadrack agents, actually contact the people they find, and see whether any of those conversations turn into real interest.