15 years in bookkeeping.
Every client loved me for the first 6 months. By year two? At each other's throats.


I ran a fractional bookkeeping business. In the beginning? I was great. New clients, 
new systems, new problems. But once the novelty wore off—same reconciliations, 
same reports, same emails asking for the same missing documents.
I thought the problem was the clients. The work. Me.
Then I got diagnosed with ADHD.
Turns out my brain runs on uniqueness. I need to feel like what I'm doing is special, 
different, mine. Repetition kills that. So I stopped fighting my wiring and started 
building around it.
Now I build AI tools that solve real problems for real businesses—starting with the 
industry I know cold.
First up: OnboardLy. It kills the document chaos that happens before bookkeepers 
can even start their real work. Because the worst part of a bookkeeper's job isn't the 
bookkeeping—it's hunting down missing bank statements at 11pm.
I work in 10-week sprints. Build, market, launch, rotate. It's the only way my brain lets 
me finish anything.
My white whale? The search tool Ezra Klein said newspapers need. Ask a question, 
get a syllabus—not a list of links, but a structured path through the archives with every 
source at your fingertips. I'm the one building it.
I'm documenting the whole journey here. Wins, failures, everything in between.
Fractional bookkeeper drowning in client onboarding? Let's talk.
Building software for an industry you've never worked in? I have thoughts.

It took me 15 years to realize: I loved accounting theory. I hated accounting work.
I ran a fractional bookkeeping business. Good clients. Steady revenue. But every month felt like Groundhog Day.
Same reconciliations. Same reports. Same emails asking for the same missing documents.
I thought something was wrong with me. Why couldn't I just be content?
Then I got diagnosed with ADHD at 45.
Turns out my brain runs on three things: novelty, interest, and urgency. Repetition isn't on the list.
So I stopped fighting my wiring.
Now I build AI tools in 10-week sprints. Build, market, launch, rotate. Always something new to conquer.
ADHD isn't a limitation for entrepreneurship. It's a design constraint. Build around it.
Anyone else out there who finally stopped fighting their brain and started working with it?

Tim O'Rourke 🧮

You want to get your groove back, right?   That feeling that you really enjoy walking in the door of your restaurant, that you want to rush through your opening activities so that you can start welcoming guests for lunch ASAP?  I'm here to help you get that back.

Why?  It all comes down to a dollar.   

When I was 12 years old my parents sold their restaurant for a dollar.  Why, because it was pushing them into bankruptcy and it was the only way they could get out from under it all.  

Fast forward 25 years and I realized why.  I was sitting in an accounting class and realized that my parents didn't get the business side of running a restaurant.  They didn't know all the stuff I was learning in business school.

Today I focus on ensuring that my clients are on sound financial footing (and do not have to sell their restaurant for $1!) by providing accounting services strictly to restaurants.  That includes bookkeeping, financial statement preparation and review, and payroll services.  For our just the right client (you, perhaps?) we even throw in cost and profit center analysis.   

Just imagine the stress of having to get payroll done, A/P out the door, and knowing there is enough money to pay for those things, and yourself!  That is the service I offer my clients. 

Shoot me a DM.  

I’ll send you a free report on how to close the three biggest holes I find on most P&Ls.  

Then you can get that feeling back that has been slowly slipping away.

Former Bookkeeper → Building AI Tools for the Industries I’ve Worked In

Your company is sitting on 15 years of answers it can't access.
That's not a data problem. It's a retrieval problem.
Here's what RAG—Retrieval Augmented Generation—actually does:
Your company has 15 years of contracts, policies, SOPs, and tribal knowledge rotting in SharePoint. The people who wrote them are gone. Nobody knows it exists. You're paying for expertise you can't access.
RAG doesn't generate fluff. It retrieves your actual information and gives it back to you when you need it. Cited. Sourced. Accurate.
Not another chatbot that sounds confident while making things up.
This is the difference between "AI-powered" and "actually useful."
Your competitors are figuring this out. The ones who get there first don't just win on efficiency—they win because their people can access 15 years of accumulated knowledge in 15 seconds.
That's why I'm building these apps. Not because AI is trendy. Because this actually solves something.

I'm teaching myself RAG by building apps that shouldn't exist yet.

Yesterday: A Q&A system over NASA mission transcripts. Today: A search tool for 160 years of New York Times archives.

Here's what I learned:

Retrieval quality is the hard part.

I asked: "What happened to Lincoln?"

It returned five articles from April 1865 — the assassination, the funeral, the manhunt for Booth, and an unexpected piece about Booth's famous actor brother grieving what his sibling had done.

That last one? I didn't ask for it. The app surfaced it anyway.

✅ What I built today:
• 1,366 articles from NYT Archive API (April 1865)
• Embeddings stored in Supabase with pgvector
• HNSW index for vector similarity search
• Semantic search returning ranked results with original NYT links

🔧 What I'm still figuring out:

Retrieval layer:
• Query expansion — vague questions need rewriting before search
• Hybrid search — combining keyword + semantic matching
• Reranking — using an LLM to re-score results after retrieval

Database/architecture:
• When pgvector is enough vs. when you need Pinecone, Weaviate, Qdrant, or Chroma
• What workloads each is optimized for (scale? hybrid search? complex filtering? local prototyping?)
• IVFFlat vs HNSW indexing tradeoffs
• Latency at scale (1,366 articles is easy — what about 10 million?)
• Metadata filtering without killing performance

The "magic" layer:
• Connection discovery — surfacing articles the user didn't know to ask about
• Entity extraction across time periods
• Knowledge graphs vs. embedding clusters

That connection discovery piece is the hard part. It's what separates a search tool from something actually useful.

I don't fully know how to do it yet.

So I'm learning. Building. Breaking things. Talking to people who've done this at scale.

I'm asking for help with two things:

1. Know someone at the New York Times or connected to Ezra Klein? I'm building this specifically for media archives and would love an introduction.

2. Work on production RAG systems or vector databases? I want to learn from people who've solved these problems at scale. If you're building RAG apps, work at Pinecone/Weaviate/Chroma/Qdrant, or have deployed retrieval systems in production — let's talk.

If you can help with either, comment or DM me.

Building in public. Learning as I go.

The Grind Has Changed—And Not for the Better

In 2012, I joined Fizzle.co. Everyone was there to make money, but we genuinely helped each other. If I asked a question, 5-10 people jumped in to help the next day.

In 2023, I found something similar on LinkedIn—a small group starting businesses and supporting each other. The last time I talked to the group leader, he was telling me about ChatGPT and how it would change everything.

Then I went offline. Until a few weeks ago.

Turns out he was right—though not in the ways any of us expected.
I tried to recreate what worked before. Joined a Skool community. Started posting on LinkedIn. Did what I've always done to build relationships online.
Fizzle? Sold to a company that shut it down. The URL now redirects to a landing page selling LLC filing services. A community that helped hundreds of people, reduced to a lead gen funnel. Everything else online seems to have done the same.

What I've noticed:

On Skool, no one responds for days. After 6-8 weeks, I've gotten maybe 2-3 helpful answers total.

On LinkedIn, I sent ~110 InMails about my AI tools, asked to chat, tried to connect. One helpful reply, ~10 "not interested," 99 crickets. The one person who replied said I was the fourth to contact him about a free AI agent—and it was January 7th. He gets requests to beta test AI apps daily.

But when I reply to a sales InMail? Response in minutes. People are there. They're listening. They're just not interested in building real connections.
This feels like a broader shift.

My most recent employer called himself "the Elon Musk of accounting" and pushed people to grind harder, automate everything, and discouraged talking. He chastised three of us for spending 10 minutes total making sure shareholder letters printed correctly. No meetings. Ever. Turnover hit 50%+ in 2025.

Everyone entrepreneurial seems focused solely on the next sale—like PE firms that squeeze "efficiency" out of acquisitions, hollow them out, and flip them.

Here's the kicker:

I used AI to write a LinkedIn post, scheduled it, and the next post in my feed had the exact same hook. We're all drawing from the same well. When everyone's optimizing, no one stands out.

What happens when the next pandemic hits and supply chains collapse? If your supplier relationships are just automated emails and lowest-bid contracts, who picks up the phone for you? You can't automate trust.
We've traded connection for optimization theater. Communities that ran on mutual support now run on engagement metrics. We've hollowed out the human element and are surprised when nothing's left.

My question:

Do you like this? LinkedIn being 95% AI content and N8N workflows given away for a comment?

If not, how do we change it?

Connect if you want to bounce ideas about building a business without grinding so hard you can't talk to anyone 8am-6pm. I'm also an accountant—tax season questions welcome.

taplio

Tim O'Rourke's Linkedin Analytics

Tim O'Rourke

Performance ScoreTim O'Rourke's LinkedIn Performance

What is Tim talking about?

Who is engaging with Tim

Tim O'Rourke's Best Posts (last 30 days)

Ready to grow your LinkedIn presence?