Building Loamly in Public: 3,000 Commits, 8 Users, and Counting

I started building Loamly in August 2025 while working a full-time job at FELFEL in Zurich. Four months and nearly 3,000 commits later, I launched on Hacker News and Product Hunt on December 30-31. I got 2 free signups.

That's it. Two.

This is the unfiltered story of what building a SaaS solo actually looks like—not the highlight reel.

Why I Started: The Problem I Couldn't Ignore

In mid-2025, I noticed something strange in my analytics.

Traffic patterns were shifting. "Direct" traffic was growing faster than anything I was doing could explain. Branded searches increased without corresponding campaigns. Users mentioned finding me through "search" but couldn't recall specific queries.

The culprit became obvious: AI.

ChatGPT, Claude, Perplexity—these platforms were becoming primary discovery channels. But my analytics tools (GA4, Mixpanel, traditional stuff) were completely blind to it. They couldn't tell me:

• What AI platforms were saying about my brand
• How much traffic actually came from AI recommendations
• Whether AI mentions were positive, negative, or neutral
• How my AI visibility compared to competitors

I looked for tools that solved this. There wasn't anything that really did it well. So I started building.

Month 1: The Foundation (August 2025)

Reality check: Building a SaaS while employed full-time means nights and weekends only. I estimated 15-20 hours per week of actual coding time.

I made a critical early decision: use a professional boilerplate.

As Two Cents Software's solo founder guide puts it: "Combined with AI coding tools, this approach typically delivers a launched MVP in 8-10 weeks of part-time work."

I started with:

• Next.js 15 with App Router
• Supabase for auth and database
• Tailwind + shadcn/ui for design
• Vercel for deployment

The first month was all infrastructure. Auth flows, workspace setup, basic navigation. Nothing exciting. No AI detection yet.

Reality: Zero users, zero revenue, lots of boilerplate setup.

The unglamorous truth about building in public: most of the early work is plumbing.

Month 2: First AI Detection (September 2025)

This is where it got interesting.

I built the first version of AI traffic detection. The approach:

1. Parse incoming traffic referrers
2. Match against known AI platform patterns (chat.openai.com, perplexity.ai, etc.)
3. Store and visualize the data

The initial detection was basic—just looking for referrer strings. But it worked. I could see actual ChatGPT traffic hitting websites.

The first breakthrough: A test site I ran showed 3% of traffic coming from AI sources. Traffic that would have been invisible in traditional analytics.

3% might not sound like much. But for some sites, that's the difference between understanding your growth and being completely confused.

The first mistake: I over-engineered the dashboard. Spent two weeks building visualizations that nobody needed yet. Should have shipped faster.

Reality: Still just me testing on my own sites. No external users yet.

Month 3: The Pivot (October 2025)

October was humbling.

Beta users told me the traffic detection was interesting, but what they really wanted was to see what AI was saying about them. Not just how much traffic—but the actual content of AI responses.

This meant building something completely different: an AI querying system.

Instead of just tracking traffic:

• Query ChatGPT, Claude, Perplexity, and Gemini
• Parse their responses for brand mentions
• Track sentiment and positioning
• Compare against competitors

I essentially had to throw away a month of work and rebuild around the new core.

Lesson learned: "Solve a real problem" isn't enough. You need to solve the right version of the problem. My early users showed me I was solving for the symptom (traffic detection) when they wanted insight into the cause (AI visibility).

Reality: One person signed up in September. Still mostly building in isolation.

Month 4: The GEO Score (November 2025)

With the AI querying system working, I needed a way to communicate results clearly.

Enter the GEO Score.

The idea: a single number (0-100) that captures your Generative Engine Optimization performance. Like a PageSpeed score, but for AI visibility.

Building the GEO score meant:

• Defining what factors matter (content clarity, schema markup, E-E-A-T signals)
• Building scoring algorithms for each factor
• Weighting and combining into one metric
• Making it actionable with recommendations

The hardest part: Getting the weights right. Initial versions produced weird results—sites with obvious AI visibility issues scored high, while well-optimized sites scored low.

The fix came from analyzing actual AI citations. What patterns correlated with getting mentioned by ChatGPT? That data informed the scoring model.

Another mistake: I tried to make the score too comprehensive. First version had 15 factors. Trimmed it to 6 core factors that actually matter.

Reality: Second signup happened in November. Two users total. Still no revenue.

Month 5: Public Launch Prep (December 2025)

December was about polish. The product worked. Now it needed to look professional.

Tasks completed:

• Complete UI redesign with consistent design system
• Landing pages for SEO/GEO keywords
• Documentation site
• Pricing page
• Legal pages (privacy, terms, GDPR)
• Email system for reports

The PageSpeed grind: I spent an embarrassing amount of time optimizing PageSpeed scores. Target was 100/100. Achieved it. Worth it? Probably not, but it felt good.

SEO and GEO optimization for the marketing site itself became a mini-project:

• JSON-LD schema for Organization, WebApplication, FAQPage
• Comprehensive meta tags
• Sitemap with proper priorities
• robots.txt with AI crawler permissions

December launch:

• Launched on HN and Product Hunt on Dec 30-31
• Posted on Indie Hackers and dev.to
• Two more signups in December (4 total users)
• Zero paying customers yet

Where I Am Now

Let me be brutally honest.

The numbers as of January 11:

• Total signups: 8 users
• Paying customers: 2 (both on Grow tier)
• Free users: 6
• Total commits: ~3,000

Four more people signed up in January. Two of them converted to paid. That's a 25% conversion rate from signup to paid, which is good. But 8 total users after a launch is... humbling.

What I'm learning:

1. Distribution is everything. I built a product in stealth mode for 4 months. I don't have a network. I don't know VCs or "10 CEOs to talk to this week." I work a full-time job. The playbooks that assume you have time and connections don't work for me.

2. The product works. The two paying customers are actively using it. The /check tool generates real insights. The problem isn't product-market fit—it's that nobody knows Loamly exists.

3. Cold outreach is the answer. I have Clay.com credits from my day job and 20 email accounts warming up in Instantly.ai. The plan is to pre-generate AI visibility reports for prospects and email them. Show value first, then ask for the sale.

The Real Numbers

Metric	Reality
Time building	4-5 months, nights and weekends
Code commits	~3,000
Total signups	8
Paying customers	2
Revenue	Not covering infra costs yet
Infrastructure cost	~$200/month (Vercel, Supabase, Upstash, APIs)
Runway	Infinite (I have a job)

This isn't a hockey stick growth story. This isn't even a growth story yet. It's a "I built something real, now I need people to find it" story.

What I Got Wrong

Over-engineering early. My first instinct was to build things "properly" from the start. Database schemas that could handle millions of users. Authentication flows for enterprise use cases. Features nobody asked for.

The truth: ship faster, fix later. You don't need perfect architecture when you have 5 users.

Building in isolation. For the first two months, I barely talked to potential users. Just built what I thought they wanted. When I finally showed it to people, they wanted something different.

As Startuups' solo founder research notes: "Talk to every user. In the early days, every user conversation is gold."

Underpricing. My initial pricing was too low. I was afraid nobody would pay. But the people who actually need this product are willing to pay for it. Wish I'd started higher.

Ignoring distribution. I spent 90% of time building, 10% on marketing. The successful solo founders do 50/50. I'm correcting this now.

What I Got Right

Solving my own problem. I genuinely needed this product. That meant I understood the use case deeply and had infinite patience for the frustrating parts.

Open source commitment. Making the core open source built trust and differentiated from closed alternatives. People can see exactly how the detection works.

Building vertically. I finished one workflow completely before starting another. Each feature is usable on its own, not half-built.

Staying focused. Dozens of feature requests. Said no to most. The product does AI visibility analytics. That's it.

The Mental Game

Nobody talks about this enough: building alone is psychologically hard.

The isolation: No coworkers to celebrate wins or commiserate over bugs. You fixed a critical issue at 2 AM? Nobody knows or cares.

The doubt: Every day brings moments of "this is stupid, nobody needs this." The only cure is user feedback—real people using your product and finding value.

The comparison trap: Other founders sharing $50k MRR screenshots while you're at $0. The antidote is remembering their journey also took years, and the highlights reel isn't the full story.

The Bootstrapped Founder has good content on the emotional side. Building in public doesn't mean hiding the hard parts.

What's Next

Q1 2026 priorities:

1. Better attribution. Help users connect AI visibility to actual business outcomes.
2. Competitor benchmarking. Make comparison easier and more actionable.
3. Content distribution. More blog posts, more SEO/GEO content, more presence.
4. Community building. Connect users with each other.

The goal isn't venture-scale growth. It's building something people genuinely need, that works well, that I can sustain solo.

If Loamly never gets bigger than a profitable one-person business, that's a success. The point was never to raise money or get acquired. The point was to solve a problem I cared about.

Building in Public: Why Bother?

Some people ask why I share this stuff. What's the point of documenting failures and revenue numbers?

Accountability. Publicly stating goals makes me more likely to hit them.

Feedback. People who follow along often have useful suggestions.

Marketing. Every post about building Loamly is also a post about AI visibility. Content begets content.

Helping others. If someone considering the solo founder path reads this and gets useful information, that's valuable.

The downside is vulnerability. Sharing revenue when it's low, admitting mistakes publicly, having your commits visible for judgment. Not everyone is comfortable with that.

But I think transparency builds more trust than polished perfection. You can see exactly how Loamly was built. Nothing hidden.

The Honest Summary

Four months in, here's where I stand:

• Product: Works. Genuinely useful. The /check tool is solid.
• Users: 8 total. 2 paying. Not enough.
• Revenue: Under $100. Not covering costs.
• Personal cost: Hundreds of hours, lots of 2 AM sessions.
• Biggest mistake: Building too long before launching.
• Outlook: I'm just getting started.

Building a SaaS solo while working full-time isn't romantic. It's coding until midnight, then waking up at 6 for your real job. It's your partner asking why you're still on the laptop. It's launching and hearing crickets.

But I built something. It works. Two people are paying for it. And I have a plan for the next phase: stop building in isolation and start putting Loamly in front of people who need it.

If you're considering the solo founder path, here's my honest advice: launch earlier than you think you should. Talk to people even when it feels awkward. And don't compare your Day 1 to someone else's Year 3.

---

Want to see what Loamly does? Try the free AI visibility check and get your GEO score. No signup required.