Humanlog started as a side project and turned into a real startup almost by accident. The product was a localhost-first observability tool - a way for developers to understand their logs, traces, and metrics locally before sending anything to a hosted platform. It ran out of Seoul for about a year and a half with a loose crew of people who signed up for an experiment and ended up building something with real global traction. This is a retrospective on what happened, what we learned, and who did the work.
If you're hiring or looking for collaborators, skip to where everyone is now - talented people came out of this and they're all available or doing interesting things.
The Incubation Phase
I (Antoine) had been running a technical incubation engine on the side - a space to explore ideas, prototype tools, and work with talented engineers on things that interested me. Humanlog grew out of that. The original product was a small log viewer. The idea was simple: developers should be able to see what their systems are doing on their own machine before committing to a vendor.
In late 2024, the incubation started to get serious. Seunggon came on as a freelance backend engineer in September 2024 - he'd been at a small Korean company and was looking for something different. Yejee joined a month later on the frontend. She'd been working as a frontend engineer at startups for a few years by the time we brought her on. Leonardo, a longtime collaborator, was never formally on the team but was a constant presence - a sounding board, a feedback loop between me and Yejee, and an early participant in our conference outreach. He was always in the room even when he wasn't on the payroll.
At that point it was a rag-tag group of people orbiting an incubation project. Nobody was building a company. We were building a tool and seeing if anyone cared.
Early 2025: The Side Project Becomes a Startup
Yejee went full-time in early 2025 as our product engineer, and that's when the incubation turned into something else entirely.
She was the first person fully dedicated to humanlog, and what happened next wasn't planned. The product needed UX research - she designed research plans and started interviewing users. It needed product direction - she synthesized feedback and decided what to build. It needed an office - she found one and set it up. It needed someone to handle the bank, vendors, and Korean-language business operations - she took care of it. It needed marketing - she invented that too.
What started as "Yejee is our freelance frontend engineer" turned into Yejee basically running the company. She took my small log viewer prototype and iterated it into a multitenant logging and tracing observability platform with real users. She was the real founder in everything but title - she just happened to arrive through the side door of a freelance frontend gig.
Seunggon continued freelancing on the backend through this period, building the Go API services and keeping the infrastructure moving. Between the two of them, they were doing most of the actual engineering - I contributed architecture direction and code part-time. Leonardo kept showing up - at planning conversations, at early conference prep, as the person you'd bounce an idea off before committing to it. Leonardo, Yejee, and I had a running feedback loop that shaped the product as much as any formal process did.
Meanwhile, I was watching my serious side project accidentally become a startup that needed a full-time founder. It had real operations now. Real users. Real responsibilities. And I was trying to support all of it while carrying serious professional commitments elsewhere.
Mid-2025: The Conference Circuit
For me, conferences were a way to mix personal interests with professional ones - travel to a new city, attend something interesting, and see if our product resonated with real people. Our rag-tag team would show up somewhere new, meet with potential users, do demos, collect feedback, run marketing, and network with industry players. No booth. No budget. Bootstrapped and scrappy.
What we didn't expect was how much we'd learn at each stop, and how fast the product direction would evolve because of it.
FOSSASIA was our first real outing. The folks there generously gave us a lightning talk spot, and a comment from the crowd gave us the insight that changed everything: people didn't just want pretty searchable logs - they wanted OpenTelemetry. Our guerrilla marketing was rough, but we made some friends.
KubeCon China (Hong Kong) taught us that people wanted tracing, metrics, dashboards - the full stack, not just logs. We had better marketing by then and learned that what we really needed was to book demos in advance.
KubeCon Japan (Tokyo) was where we ran our first effective demos. The signal was clear: people wanted all of this, and they wanted LLMs integrated into it.
KubeCon India (Hyderabad) was the culmination. People didn't want to write queries at all - they wanted LLMs to do everything. Yejee had been tracking this pattern since Tokyo and pushed us to stop investing in our custom query languages entirely. She was right. She made a similar call on our auth stack - WorkOS was causing onboarding friction, so she led the migration to better-auth in-house. By this point, we had perfected the guerrilla approach and ran a full demo schedule with targeted marketing across physical print media and a much more prepared logistics effort. Still guerrilla - still bootstrapped and scrappy - but dialed in. Yejee ran 30 user interviews in two days.
At every conference, Yejee was the engine. She came up with the idea of placing our flyers in the bathrooms at KubeCon - the one place every single attendee visits. Nobody else was doing it. The effect was immediate: our demo and research calendar filled up completely. She ran the demos herself, talked to users, collected feedback, and wrote structured reports afterward so we could act on what we heard. The whole loop - design the campaign, execute it on the ground, conduct the demos, capture the feedback, write the report, feed it back into product - she owned end to end. She came back from each conference with structured notes and recordings, broke them down into direction changes, bugs, and quick wins, and distributed the work across the team with deadlines attached. When a pitch wasn't landing on the floor, she'd read the room and shift our approach mid-conversation. Leonardo helped with early outreach and was in the mix throughout.
Beyond product feedback, we managed to do something unexpected: we connected international companies with the Seoul tech community. We helped bring ClickHouse to Seoul for their first official meetup in Korea, in collaboration with Dev Korea - thanks Florian. Being small and present and genuinely caring about the ecosystem turned out to be its own kind of leverage.
What we learned about guerrilla dev tools marketing
Being small, nimble, present, and consistent is very effective. We showed up everywhere, talked to everyone, and followed up. The effect compounds over time - people start recognizing you, referring you, inviting you to things.
Mix feedback collection with marketing. We didn't treat conferences as "get leads" events. Every demo was also a research session. Every conversation taught us something. This made our presence feel genuine rather than salesy, and people were more willing to engage.
Creative placement beats expensive booths. The bathroom flyer trick generated more qualified conversations than most booths get. Pattern interruption matters - everyone has seen a hundred booths, but nobody expects marketing material next to the soap dispenser.
Write it down immediately. Yejee's post-conference reports were critical. Conference learnings evaporate within a week if you don't capture them. She documented not just what people said, but patterns across conversations - common objections, recurring feature requests, which messages landed. This became our product roadmap.
Along the way we met a lot of people who were incredibly helpful - gave us great feedback, connected us to more people, visited us in Seoul. We visited them back. The conference circuit built a real network, not just a pipeline.
The Product
The product evolved in lockstep with what we heard at conferences. It started as a log parser with a search engine - pipe your output in, get structured, searchable, human-readable logs. Then it became a localhost OTLP endpoint that could ingest logs and traces via OpenTelemetry. Then we added dashboards and alerts. Then it grew into a localhost-to-production workflow - use it locally while you build, flip a switch and send the same telemetry to a hosted backend when you're ready. By the end, we were prototyping LLM-powered querying and analysis that hadn't left the lab yet but pointed at where the product wanted to go.
What Yejee built on the frontend was the humanlog.io web application - the interface for all of this. Not a Grafana fork, not a template, something purpose-built. She iterated it through every stage of the product's evolution: from simple log viewer to full platform with gRPC-like APIs for querying, user management, multi-tenant organizations, and feature flags. It pulled telemetry data from ClickHouse, handled Stripe billing, managed authentication through multiple migrations, and supported real-time data streaming.
On the backend, Seunggon built and maintained the API services in Go - the query layer, user and org management, the connective tissue between the frontend and the data infrastructure. He became the person who knew where DuckDB and ClickHouse agreed and where they didn't - every analytical function, every edge case - and managed to make both backends look identical to end users. That's unglamorous work that only matters until it breaks, which it never did. I worked part-time across the whole business - architecture, external relationships, product strategy, and some code - while Seunggon carried most of the backend engineering and infrastructure work day-to-day.
The hosted product ingested OpenTelemetry data - logs, traces, and metrics - stored in ClickHouse for fast querying. The architecture was real. The product worked. People signed up from around the world despite minimal marketing and an obviously incomplete MVP.
October 2025: The Full Team
By fall, we were ready to grow. Yejee had long outgrown her product engineer title - she was our head of product in practice, so we made it official. Seunggon moved from freelance to a full-time role. Daeun joined as our first dedicated UX/UI designer - she'd been a product designer at an AI SaaS company and a design manager for UI/UX and graphic design before this, and humanlog gave her a new canvas.
Daeun's arrival mattered because Yejee had been doing all product design work for almost a year. Daeun knew nothing about dev tools or observability when she started. She fixed that fast - interviewed our team and outside engineers, researched how existing products worked and where ours would fit into real workflows, and developed genuine developer-empathy for a domain that's notoriously hard to design for. She turned that into a complete product design for what humanlog should become, paired it with strong graphic design, and managed external design freelancers alongside her own work. Having her meant Yejee could focus more on engineering and product ownership.
For a brief window, we had a full team: Yejee on product and frontend, Seunggon on backend, Daeun on design, Leonardo as the ever-present advisor, and me part-time on architecture and the business side. It felt like the thing was about to hit its stride.
November 2025: The Music Stops
And then we started to see it. A trickle of signups from around the world - consistent, unprompted, organic. AI companies signing up for our localhost-first tool to monitor their LLM calls and agentic applications. It showed up in September and October, and it smelled like the signal you wait for: real demand, from a market that was exploding, finding us without us chasing it. Everything pointed to scaling up.
But my bandwidth had always been part-time, and even that was getting crowded. Family emergencies demanded more and more of my attention. I tried to find someone who could step into the full-time leadership role and carry the thing forward, but time ran out before I could. I had to tell the team that the experiment would have to come to a stop.
Rather than let things degrade slowly, we decided to pause where we were. Take the wins. Let everyone find a good landing spot. The wind-down was announced with three months of paid notice and structured so everyone came out whole. The company was kept as a shell for a potential restart.
What Everyone Did Next
Seunggon's arc through humanlog was remarkable. He came in as a backend engineer from a small Korean company, joined a loose incubation project led by a foreigner building developer tools as a side project, and watched it turn into a real startup. During the notice period, he didn't wind down - he ramped up. He took a half-disassembled 3090, rebuilt it into an ML workstation, and started fine-tuning a Qwen3 model to translate natural language observability queries into our custom KustoQL implementation - teaching himself the entire post-training pipeline from scratch. He went from backend engineer to nascent ML engineer over the course of a few months, and landed a position at Trillion Labs, the top Korean AI lab. The trajectory - small company to scrappy startup to AI lab - is the kind of career acceleration that early-stage work can produce when someone is hungry for it.
Daeun brought product design and design management experience from an AI SaaS company and turned it into something deeper at humanlog. In her time at humanlog, she built a complete UX research project and a full product design - not mockups, but a structured research practice feeding into a real design system. She proved she could do the work, and she did it fast.
Yejee's story is the most striking. She took a freelance frontend gig at a side project and turned it into running a startup. She iterated a small log viewer into a multitenant observability platform. She conducted the UX research that shaped the product, designed the guerrilla marketing that filled the conference calendar, ran the office, handled the business operations, supervised freelancers, and represented the company at KubeCon. She had instincts that most engineers don't have: how to read a room, how to make someone feel attended to, how to turn a hallway conversation into a relationship. She was a cofounder in everything but title.
What I Learned
I've learned many things from this, but the most important ones are personal.
Things take a shape of their own. Sleep-walking into being an unavailable, wrong-timing-to-be CEO of a startup that was actually run by a self-organizing team - well, it ends up being not great for the team. I realized this too late and should have stepped away from the leadership role sooner and let someone else take what I was never going to be able to fulfill. That's on me.
You absolutely need cofounders. Full dedication. Some businesses inherently need funding to scale once the signals are found, and finding yourself unable to make the commitment to scale is a death sentence for the endeavor - no matter how strong the signals are.
Guerrilla marketing works. You don't need a booth or a budget - you need creativity, presence, and follow-through.
Hire talented people and they will surprise and delight you. And make sure you have a good time - eat good food, drink good coffee, have good conversations, and make lifelong friends along the way. Know when to stop and capture your learnings.
On the product side: Localhost-first observability has real demand - people signed up organically from around the world. But the demand is for a workflow that leads to a hosted product. The localhost tool alone isn't something people pay for.
On building from Seoul: Korea is a viable base for building a global developer tool. We went to international conferences, got users worldwide, built real relationships with the global community. The goal was always to sell abroad, not locally. But we learned that the language and credibility gap cuts both ways - foreigners in Korea struggle to partner with local companies, and Korean dev tools struggle to break out of an insular market. Building global from Seoul is possible, but bridging that gap in either direction takes deliberate effort.
On what matters: The best thing a startup can produce is people. Everyone who touched humanlog came out of it with more skills, more confidence, and a better trajectory than they had going in. That might matter more than the product.
What About Humanlog?
What started as a log prettifier CLI became a log search engine, then a full observability platform. Through the experiment, humanlog was in a constant state of flux - destruction and reconstruction, pivoting and rebuilding, sometimes week to week. We thank our users for their patience with that.
Going forward, humanlog is returning to what it was at the start: a log parser and prettifier. The observability tool is being spun out as minitape and returning to side-project status. The project is being open-sourced, including the frontend - though the frontend isn't expected to be reusable and isn't licensed as such.
A heads up: the MVP is not currently open for signups and the API is partially shut down. If time permits, the hosted observability tool may come back under the minitape name in the future, as Antoine continues to need an observability swiss knife for his own work. But no promises and no timeline.
👋 Where Everyone Is Now
Yejee (GitHub) is exploring opportunities as a product manager, project manager, or product engineer - in hospitality tech, SaaS, data visualization, or anywhere that needs someone who can own a product end to end. If you need someone who can cross the gap between customers and implementation, someone who can ship code lightning fast, and someone who's already proven she can run the show: she's open to remote work or flexible hybrid arrangements in Seoul. Reach out to her.
Daeun is turning her humanlog experience into a UX research and product design studio, and she's taking clients. If you want a capable product designer who will learn your domain, your product, and your industry and turn your ideas into a solid product design - contact her. Limited spots.
Seunggon (GitHub) is now at Trillion Labs, the top AI lab in Korea. Through humanlog and beyond, he became an expert in ClickHouse, DuckDB, parsing, transpiling, ConnectRPC, and the full stack of data infrastructure. He's exactly where he should be.
Antoine (GitHub) is focusing on his kids, a demanding day job, and scaling down project incubation to advising and angel investments into technical founders targeting global markets from Seoul. Reach out for coffee.
Leonardo (GitHub) is building Foliofox and is part of the Antler Korea batch. The guy who was always in the room is now building his own room.
Humanlog is paused, not dead. The shell is still there. The learnings are real. And the people who built it are ready for what's next.