Are Hackathons for Children?
The full story of how we rebuilt our product six weeks before our first hackathon — and why you should do the same.
This is the full, uncut version. If you want just the hackathon highlights, check out the FOMO Observer edition.
When I was inviting people to our hackathon, one guy told me he wasn’t coming because it was going to be a bunch of children. I said okay with a laughing emoji and moved on. But it made me realize that most people have no idea what a hackathon actually is, who shows up, or what it can do for a company. So let me tell you the whole story of ours. The insane rebuild, the chaos, the wins, and why organizing a hackathon was the single best thing we did this year (I know it’s only March). And I hope that after reading this, you’ll stop whatever you’re currently doing and go participate in one. Or even better, organize it yourself and invite us to co-organize it with you.
How It All Started
I met Jimmy, Monomer Bio’s CEO, last September. I sent him a cold DM on LinkedIn and just said, “Hi, we’re building Elnora, an agent for preclinical lab work.” He was like, “Yeah, sounds cool, let’s have a call.” That’s when I met him the first time.
It was immediately super clear that Elnora and robotic labs go together. We are extremely complementary: Elnora generates lab protocols for preclinical work, but you need someone to execute those protocols. That’s where robotic labs come in. Way faster, way more efficient, and you can really close that loop if you have a robotic lab.
Then Monomer Bio organized their first AI scientist cell culture hackathon with other AI partners at the end of October. The whole LinkedIn was full of posts and I felt the biggest FOMO ever. First, I wanted to go and participate as a builder myself, but I was occupied with something else and couldn’t make it.
But then I thought, wait a minute, we should be that AI partner. I want to be that AI partner for Monomer Bio. So I reached out to their team and did a demo for the whole team, Carter, Turner, Jimmy, and others, and showed them what Elnora can do. They were like, “Yeah, let’s do it. Sounds great. You’re our next partner for the hackathon.”
On social media, it almost looked like it was my hackathon, but that’s definitely not the case. It was Monomer Bio’s hackathon, and we were proud to become an integral part of it. Michael Raspuzzi from Worldwide Studios was the main organizer who brought everyone together — Monomer Bio, Elnora, Luis Villa from Bay Area Lab Automators, and partners like Cephla, Opentrons, Novel Bio, SF Hardware Meetup, and PyLabRobot. Without Michael, this event simply wouldn’t have happened the way it did. By the time the hackathon came around, Elnora had become deeply woven into the experience — but this was always a very strong community effort.
We Decided to Rebuild Everything
We agreed on a date: March 14th and 15th, 2026. But from December through the end of January, we were deep in SOC 2 and ISO audits and we couldn’t really do anything else, just focus on our clients and the compliance work.
Then 6 weeks before the hackathon, we looked at our product and knew it could be way better. The first version of Elnora was an agentic workflow, but not a real agent. It was kind of inside the box. I wanted Elnora to be this super powerful agent with thousands of tools that can do anything a scientist needs, giving them actual superpowers, and our old architecture didn’t allow that. So I told Risto, “Let’s build a completely new version.” He looked at me and said, “We have literally a month and a half, and you want to rebuild everything?” Yes, a totally new version. He thought I had lost my mind, and he was right, but we did it anyway.
A hundred percent of our code is written by Claude Code, every single line. But Claude doesn’t build by itself. Risto’s (or any human’s) role in the process is guiding the research, making the key architecture decisions, planning the entire build sequence, and reviewing every piece of output to make sure we maintain full control over the product. Claude is an incredible builder, but it needs a human who knows what they want, why they want it, and can tell when something is off. It’s not vibe coding — it’s engineering with an AI partner under constant human supervision. A lot of magic does happen, but the amount of building was still tremendous.
The first two, even three weeks were heavily focused on research and planning the new architecture, though Risto was already writing code during that time too. But the heavy building really kicked into gear in the last three weeks, and that’s when the pressure became insane.
Everything Broke
Two weeks before the hackathon, Risto was up every day until 3 a.m. just building and building, then waking up at 6.45 in the morning. I was up until 3 a.m. as well. Usually I don’t build stuff for the actual Elnora production because I’m still at a vibe coder stage myself and I have my own personal agent repo where I can break things without major issues, but even I was building in production this time. I was building Elnora’s CLI and MCP servers because it’s something I had done like ten times before for my own agents. One day Risto came to work literally half asleep and I told him, “Risto, you need to go to bed at 10 p.m., otherwise you will just fall asleep in front of your computer.”
About ten days before the hackathon we decided to do a mini hack with Monomer Bio to try everything out before the real event. We shipped Elnora’s new platform literally hours before the mini hack. Risto had shipped 70,000 lines of code, yes 70,000 and that’s not a typo. But it wasn’t ready yet. You sent a “hello” to Elnora and it went into the wide universe without hearing anything back. Hello, hello, hello... Nothing.
But we showed up at the mini hack anyway. I looked at Carter from Monomer Bio on Zoom and said, “Everything is broken, we just shipped a new platform and absolutely nothing works.” He was like, “Okay, that’s typical, no worries, let’s still do it. You can mock stuff, use Claude Code or whatever, let’s play it through imagining that Elnora’s agent is giving you answers.”
That was incredibly helpful because we could practice the whole flow of how a researcher asks Elnora to generate a protocol, sends it to Monomer’s robots to execute, gets results back, and then adjusts the protocol based on those results. Really trying to close that loop between the AI and the robotic lab.
After the all-day mini hackathon, it was clear. We were in big trouble. We had ten days until 50 people would all come together expecting Elnora to knock their socks off, and the agent was totally silent.
Finally, a day before the hackathon, the day I was flying to San Francisco, I woke up and Risto says, “Carmen, Elnora said hello back.” I couldn’t believe it. After nine days of silence, it actually responded with “Hello, how can I help you?”
Hackathon Day
We reached out directly to over 500 people. More than 150 registered and we had to stop the waitlist because we could only fit about 50, we already had 75 registered and another 75 on the waitlist. Out of those, about 43 people actually came for a full 24 hours of building, and the gender balance was roughly 50/50, which I was very happy about.
We were the opening act. We gathered all the people and said, “Hey, this is Elnora. Here is your API key. You can use the CLI or MCP, which I built, or the regular UI, go and knock yourself out.”
While I was registering people to the platform and sorting everyone’s API keys out, Risto was monitoring the agent responses through GreenFlash, our observability partner, as the first builders started testing Elnora out. He found several bugs like the agent truncating responses so the user could only see the top half of the protocol. He literally had to fix and deploy on the production platform at the same time when people were already on it. I wasn’t sure if we were going to make it, really. My heart wanted to jump out of my chest.
But we did. Elnora started sending out full responses and builders loved the agent. It was just incredible.
The Tracks
We had three tracks. Track A was to onboard a new cell line and use Elnora to generate a media composition that would maximize the delta, meaning the difference between how many cells you started with and how many you ended up with. We were using a bacterial cell line that’s one of the fastest growing on Earth, it doubles every 10 minutes, so you see results very quickly. The job was to use Elnora to generate lab protocols, send them to Monomer’s robotic lab, receive the results, feed them back to Elnora, and optimize. As many rounds as they could to get the best media composition.
Track B was to get a single cell on a well, which sounds trivial, but it’s not at all. You have a 96-well plate, and in every well, you want just one single cell, so you can start growing a colony; you don’t want two different cells as the starters of colonies. For this track, they used beads instead of actual cells to simplify the process.
Track C got a robotic arm and needed to program it to perform actual bench work: grab cell flasks from the incubator, place them on the bench, image them, change the media, passage the cells, and return them. Basically, automate an entire day of manual lab work.
Officially, Elnora was only part of Track A. But in the end, we had 6 teams out of 7 incorporate Elnora in their workflow, which was not something we expected at all.
The Winning Team
The winning team was a robotic arm team, and they built a whole closed loop. The robot took out the cells, imaged them, confluency was at 70%, and sent that information to Elnora. Elnora said, “Looks like cells are ready for splitting, here is a lab protocol for you to do that.” The robot goes, does the work, puts the cells back.
Then Elnora says, “Okay, let’s take another look at those cells.” The robot takes them out, images them, sends back: 5% confluence. Then Elnora says, “No, we can’t split those cells, but something seems off because we went from 70% to 5%, so we need to change something in the media.”
When I saw that demo I was just gasping for air. I could not believe it. The way they connected Elnora as an agent to the robot performing those lab protocols was just genius. It was so so so beautiful. I loved it so much. They were everyone’s favorite team, and they won.
Both Track A teams managed to increase the growth delta compared to the baseline media, which was the whole point. One of them also won the best demo award for building an app that anyone could scan a QR code and see the experiment results in real time.
One of those teams also ran Elnora and Claude side by side on the same task and Elnora kicked Claude’s ass. Claude is my favorite, don’t get me wrong, but because Elnora has so many specialized tools and skills that Claude doesn’t have, it performs way better on science tasks. That was really cool to see.
Most Scientists Don’t Code
Opening a terminal is the scariest thing for a non-coder. Yet this is where you can do the most powerful things with agents.
Many biologists who had never coded before came to the hackathon. We have a UI, but my favorite way to use Elnora is via the CLI. When biologists arrived, and I said, “Let’s start working on lab protocols,” they asked how. I explained that Elnora is a CLI tool, so you’d need another coding agent, such as Claude, Codex, or Gemini, to orchestrate it.
And people said they didn’t have a coding agent. I was shocked. How do you do your everyday job without one? I genuinely cannot remember life before coding agents, I use Claude for everything, and Elnora for science work. I just assumed everyone had Cursor, Codex, Gemini, or Claude. Totally wrong assumption. 90% of scientists at our hackathon had never coded and had no tools, let alone subscriptions.
Before I could even get them started on Elnora, I had to teach them the fundamentals first. What a terminal is, what VS Code is, what an API key is and why you treat it like a secret. I love doing this because I’m an ex-scientist myself, and I understand exactly where they were coming from. But it was a wake-up call.
Then there was the communication gap. Using agents isn’t natural to humans. When I talk to my best friend Maria I say, “Yesterday I went to a store, and I was like...” and she understands me immediately because she’s known me for 15 years, she has context. An AI agent straight out of the box doesn’t know you; it has exchanged a couple of words with you and that’s it, no experience, no context. So you have to be very clear and straightforward. That’s a skill most people haven’t developed yet.
The biggest learning from the whole hackathon was that I’m living in a bubble, and a lot of people aren’t in the same bubble. I assumed the whole world keeps up with every new AI tool. They don’t. Two and a half years ago, I was that person too, a regular scientist sitting behind the bench all day doing traditional science in academia. The initial learning curve is sharp, but after a month or two, I was already at a top level. It’s not that I’m the smartest person on earth, definitely not, but I had an amazing teacher in Risto, and I read a lot. Every day, I try to read at least one blog post about agents or AI tooling, and when I find something interesting, I try to implement it immediately. That’s the fastest way to learn.
And now it’s my personal mission to help scientists start coding. By coding, I don’t mean writing Python. I mean using English as a coding language.
What We’re Changing
The biggest product takeaway was clear: we need to eliminate the need for another coding agent to orchestrate Elnora. That was just too hard for participants. If 90% of your users don’t have a coding agent and have never used a terminal, you can’t make that a prerequisite. So that’s our number one focus now — making Elnora something anyone can use without needing to set up Claude or Cursor first.
We also got feature requests I didn’t think of myself. One person said, “Elnora, draw me a graph.” Elnora can’t draw, but it said, “I can’t do it, but I’ll give you code instead. If you copy-paste that code on that website, you’ll see my drawing.” The person did it, and it worked. I guess we are now teaching Elnora how to draw graphs.
Why You Should Organize a Hackathon
Well, after you heard my story, do you still think that hackathons are for children? Let me tell you who the people were in the room going through all this I just shared with you.
Working professionals from pharmaceutical companies with hundreds of thousands of employees, scientists who have been running labs for 20 years, AI engineers, robotics people, people who code every day, and people who have never coded before. I’m 34, and most people I see at hackathons are around my age or older. These are people with real jobs and real expertise who came because they want to experience what it feels like to have completely free hands and a free mind to create something.
What makes a hackathon special is exactly the mix. You get a bunch of people who are similar to you but also completely different at the same time, and you combine all that expertise into one team that in a regular environment would probably collapse, but in a hackathon environment is magnetic. People just pull toward each other and tackle problems together. It doesn’t matter if you’re a CEO, an associate, or an intern, everyone is mixed; there’s no space for ego, and the only rule is that there are no rules.
So why should you, as a company, organize one?
We rolled out a brand new version of Elnora days before the hackathon and gave it to real people who are not our friends, and watching how they actually use your product is worth more than months of internal testing. The day after our hackathon, our website views were 5-6 times higher than usual, all organic, all from real people. We had pharmaceutical companies in the room and the conversation immediately turned into “hey, you just used Elnora successfully, what about rolling this out at your company?” Even though we weren’t actively recruiting, people reached out saying they wanted to work for us. And by organizing together with Monomer Bio, we proved we can collaborate on shared projects with shared clients going forward.
You don’t need a massive budget or to be a big company. You need a real challenge from your work, not something made up; a venue with tables and good wifi; food and coffee for people who will stay overnight; and a group of curious people who want to build something together. Find sponsors who believe in what you’re doing. Find a partner company that complements what you do, set up a Luma event, and market it hard. People’s attention span is short so you need to cast a wide net.
And if you want to organize a bio hackathon, an AI hackathon, or anything where agents, science, and building things come together, reach out to me. I want to help you do it. I want to be there when your participants have that moment where they see something they built actually work, and they can’t believe it.
My best friend Maria texted me the other day saying that coding gives her higher dopamine than scrolling TikTok. You get addicted to it. It’s so rewarding. Go do it. If you don’t know how, participate in a hackathon, find people who know how to build agents and can teach you, and then you can do it too.
So hackathons are definitely not for children. They’re for people who are ready to build something that 24 hours ago was someone’s wildest dream.
Big thanks to our sponsors!
I want to highlight two of our hackathon sponsors — outstanding companies led by extraordinary women.
Anyone who has met me in person knows — as soon as I open my mouth, the first question is always: “Where are you from?” I still have this tremendous Estonian accent, and I am so proud to say I’m from Estonia. The follow-up is always: “So what are you doing here in America?” A big part of that story is Lisa and Plymouth.
They helped me get my O-1 visa. I met Caleb from their team, and nine weeks later I was holding a valid visa in my passport and stepping on a plane to the US. But what makes Plymouth truly remarkable is that their support didn’t end with the visa stamp. Lisa organizes dinners, events, brings amazing immigrant founders together, and makes the impossible possible. Whenever I need something, Caleb is always a call away. If you’re an immigrant founder, AI engineer, or scientist and want to build your own story here in the US — Plymouth is there for you like they’ve always been for me.
And then there’s Lauren and Cell Culture Collective, Inc. I was a bench scientist for 12 years in academia. I’ve used a tremendous amount of plastic, reagents, and materials in the lab, and there were always issues — a supplier changing a formulation, screwing up my experiments, and no one from customer support ever picking up the phone. Once I couldn’t grow HEK cells. HEK cells! These are cells that should grow on everything and everywhere — but nope, not on my dishes.
That’s exactly what Lauren and Cell Culture Collective are here for. When I reached out to ask if she could support our hackathon, it was an immediate yes. Gloves? Plastic? Reagents? And you needed it like yesterday? She’ll get it for you.
Carmen Kivisild is CEO and co-founder of Elnora AI, building AI agents for preclinical lab protocol generation in pharma and life sciences. If you want to co-organize a hackathon or talk about agents for science, reach out at carmen.kivisild@elnora.ai.
Awesome recap! And I like you designed things from the ground up for the hack. It definitely paid off as teams really enjoyed their experience with the Elnora agent
So glad you could come and join this one and great to see agents make their way onto hardware
Can’t wait for the next one!