Presented by GTM Machines
The Growth Hackathon
Winner gets a job interview or an internship.
A competitive sprint where you build AI-powered growth systems against real go-to-market problems. Solo or in a team, any stack. Winners leave with interview and internship offers, not certificates.
- Date
- Saturday, 22 August 2026
- Time
- Doors 10:40 AM · Kickoff 11:00 AM
- Venue
- Bengaluru, Karnataka
- The exact address is shared on Luma once your registration is approved.
Before you arrive
Doors at 10:40 AM. Kickoff at 11:00 AM sharp, and the clock runs whether or not you are in the room. The sprint is four hours long.
Sort out tonight
Three things to sort out tonight rather than tomorrow.
- 1
Laptop and charger
Power strips are limited, so treat a full battery as the fallback.
- 2
One coding agent subscription, active and signed in before you leave home
Claude Code, Codex, Copilot, Gemini, or Antigravity. Any one of them is fine, and you only need one. Sorting out billing or a login at 11:15 costs you a fifth of the sprint.
- 3
A phone hotspot
Sixty people running browser agents on one office wifi is the most likely way to lose an hour. Bring your own backup.
How the day runs
Worth knowing before you walk in.
Read the statements tonight and pick two
Pick two you would be happy to build. Teams form in the room right after kickoff and statements lock ten minutes later. Nobody switches after the lock.
Building solo is fine
Raise a hand at team formation and stay in your seat.
Answer the five fields in the first twenty minutes
Every entry answers the same five fields on the submission form. Write those down in the first twenty minutes, not in the last five.
Build for a stranger's input
Judges hand every finalist an input it has not seen. A demo that only works on the data you prepared will not survive that.
Every entry answers the same five fields
They are required on every submission in the room, not only on the growth-hack statement. Write them down in the first twenty minutes, not in the last five.
- 1ICP
- 2Hypothesis
- 3Channel
- 4Conversion path
- 5Success metric
The problem statements
Read them tonight and pick two you would be happy to build. Statements lock ten minutes after kickoff and nobody switches after the lock.
M1Make mem0 the default memory stack for creatorsHow can we make mem0 and memory the default stack whenever a creator (YT, X, or Insta) is creating a production-ready agent?
Written by Mem0
Build
How can we make mem0 and memory the default stack whenever a creator (YT, X, or Insta) is creating a production-ready agent?
M2Fix the benchmark LLMs cite for mem0 vs ZepLLMs cite the wrong benchmark for mem0 when asked about "mem0 vs Zep benchmark."
Written by Mem0
Build
LLMs cite the wrong benchmark for mem0 when asked about "mem0 vs Zep benchmark."
M3Get mem0 to 100k GitHub starsHow can GitHub stars be increased to 100k?
Written by Mem0
Build
How can GitHub stars be increased to 100k?
C1ICP-Specific Lead Generation EnginePick one ICP from their list. Build the system that answers: which companies should Persistence target, why do they fit, who are the decision-makers, what signal says they need voice automation, and how are they ranked.
Written by Persistence
Build
Pick one ICP from their list. Build the system that answers: which companies should Persistence target, why do they fit, who are the decision-makers, what signal says they need voice automation, and how are they ranked.
Done looks like
You enter an ICP or an industry and get back qualified, ranked companies with named decision-makers and the signal behind each.
The judge hands you
An ICP or an industry.
Demo it by
Typing the ICP and watching it discover and qualify in real time.
Watch out
Stopping at a list of companies. The "why them" and "who decides" is the actual ask. Signals worth hunting for voice AI: call-centre job posts, "call us" as the primary CTA, published support hours, IVR complaints in reviews.
C2AI-Powered Outbound Growth MachineResearch a company and a person, work out their likely pain, work out where a voice agent would help, then write the outreach.
Written by Persistence
Build
Research a company and a person, work out their likely pain, work out where a voice agent would help, then write the outreach.
Done looks like
You give it a company name and get personalised outreach grounded in something specific and true about that company.
The judge hands you
A company or a prospect name.
Demo it by
Four beats on stage: research the company, name the pain point, show where Persistence fits, generate the message.
Watch out
Personalisation that is really a merge field. The test: could the first line only have been written about that one company? Bonus if the message changes when the signal changes.
C3Growth Asset That Generates Its Own DistributionSomething a Persistence buyer would use on its own. A calculator, a benchmark, an assessment, a report generator. "How much are your calls costing you?" A voice AI readiness score. A call quality analyser. An industry benchmark.
Written by Persistence
Build
Something a Persistence buyer would use on its own. A calculator, a benchmark, an assessment, a report generator. "How much are your calls costing you?" A voice AI readiness score. A call quality analyser. An industry benchmark.
Done looks like
A working tool a stranger can use and get real value from, with a natural path to Persistence at the end of it.
The judge does
Acts as a potential customer, uses your product, and gets a useful output.
Demo it by
Handing it to a judge and letting them use it.
Watch out
This is the one statement where the thing has to be genuinely useful to someone who has never heard of Persistence. A thin lead-capture wrapper fails its own test. Build the tool first and the capture second.
C4New Distribution ChannelPick an ecosystem that already reaches Persistence's buyers — BPOs, system integrators, consulting firms, CRM platforms, contact centre providers, SaaS platforms, developer communities, industry associations — and build the mechanism that turns it into a channel.
Written by Persistence
Build
Pick an ecosystem that already reaches Persistence's buyers — BPOs, system integrators, consulting firms, CRM platforms, contact centre providers, SaaS platforms, developer communities, industry associations — and build the mechanism that turns it into a channel.
Done looks like
Something tangible. A partner discovery and scoring engine, a co-selling tool, a white-label demo, a solution configurator, a partner landing page with a lead flow.
The judge hands you
An ecosystem, or a partner name.
Demo it by
Showing how a partner would discover Persistence, understand the opportunity, use or sell it, and refer a customer.
Scope
Read this before you pick it. The full version (marketplace, onboarding, co-sell motion) is a quarter of work, not an afternoon.
Watch out
Pick one component and finish it. Partner discovery and scoring is the piece that fits a four-hour sprint.
C5Find and Execute a 30-Day Growth HackYou own growth for one ICP. Limited budget, 30 days to generate qualified pipeline. Build the first working version of the mechanism.
Written by Persistence
Build
You own growth for one ICP. Limited budget, 30 days to generate qualified pipeline. Build the first working version of the mechanism.
Done looks like
Something that runs, plus the five numbers: ICP, hypothesis, channel, conversion path, success metric.
Demo it by
Showing the experiment actually working. The agent finding prospects, the system generating demos, the loop producing shareable output, the tool capturing information.
Watch out
A campaign described on a slide scores nothing. The rubric rewards things that run. Note that the five fields are required on every submission in the room, not only this one.
B1The Demo MachineAn agent that takes a repo or a product URL and produces a finished piece of content without you steering it. A CLI walkthrough, a product video, a written explainer.
Build
An agent that takes a repo or a product URL and produces a finished piece of content without you steering it. A CLI walkthrough, a product video, a written explainer.
Done looks like
You point it at something, walk away, and come back to a file you would actually publish.
The judge hands you
A repo or a product URL.
Demo it by
Running it live on the judge's pick, then showing the output.
Watch out
"Content" is a wide word. Fix your output format in the first twenty minutes and do one format well.
B2The Second RunA system for a long-horizon job — Reddit community building, social listening — where the second run is measurably better than the first because of what it learned in the first.
Build
A system for a long-horizon job — Reddit community building, social listening — where the second run is measurably better than the first because of what it learned in the first.
Done looks like
Two runs side by side, with the improvement named as a number.
The judge hands you
A subreddit or a topic.
Demo it by
Showing run one, then run two, and saying exactly what changed and why.
Watch out
Nothing truly self-improves in four hours. Don't promise 24/7 or a scheduler; you will be taken apart in Q&A. Compete on "run two beat run one, and here is the mechanism."
B3Cold StartA system that takes a product it has never seen and finds its first 50 customers. Where they are, why them, what to say.
Build
A system that takes a product it has never seen and finds its first 50 customers. Where they are, why them, what to say.
Done looks like
A live URL goes in. Named prospects with a reason each come out.
The judge hands you
A product URL.
Demo it by
Running the whole thing on the judge's URL, live.
Watch out
Fifty is the headline number, but twenty done well beats fifty done badly, and a judge clicking a random row will find the difference. Every row carries its reason and its source.
B4Reconstruct the Competitor's Customer ListWho uses your competitor is public if you look. Case studies, job posts that name the tool, integration directories, review authors. Assemble the list and rank it by how likely each one is to switch.
Build
Who uses your competitor is public if you look. Case studies, job posts that name the tool, integration directories, review authors. Assemble the list and rank it by how likely each one is to switch.
Done looks like
Named companies using competitor X, each with the URL where you found the evidence, ranked with the switch reason.
The judge hands you
A competitor name.
Demo it by
Naming the competitor, running it, then opening three rows and showing the evidence.
Watch out
The switch-likelihood ranking is where this is won. Anyone can scrape logos off a case studies page.
B5The Adversarial BuyerAn agent that reads a company's site and pricing, becomes its hardest and most skeptical buyer, and tries to kill the deal. The output is a ranked list of the objections that go-to-market cannot currently survive.
Build
An agent that reads a company's site and pricing, becomes its hardest and most skeptical buyer, and tries to kill the deal. The output is a ranked list of the objections that go-to-market cannot currently survive.
Done looks like
Ranked objections, each tied to the specific line on the site that causes it.
The judge hands you
Their own pricing page.
Demo it by
Running it on a judge's company in front of the room.
Watch out
Don't ship a chat transcript. Rank by damage and quote the exact page line behind each objection.
D1Landing Page Optimization Through Competitive ResearchRead the competitor set, then return ranked, specific changes to your own landing page. Actual replacement copy and the reason for it.
Build
Read the competitor set, then return ranked, specific changes to your own landing page. Actual replacement copy and the reason for it.
Done looks like
A ranked list of changes, each tied to what a competitor does differently and why it matters.
The judge hands you
Their own landing page.
Demo it by
Running it on a judge's page live and reading out the top three changes.
Watch out
Generic CRO advice ("strengthen the hero", "add social proof") scores nothing. Every recommendation carries a competitor URL behind it.
D2End-to-End Analytics FunnelInstrument a funnel end to end, work out who is actually visiting, then build the GTM motion that pulls more of that kind of person in.
Build
Instrument a funnel end to end, work out who is actually visiting, then build the GTM motion that pulls more of that kind of person in.
Done looks like
The funnel instrumented, one visitor segment named with evidence, and a concrete motion to get more of them.
Demo it by
Showing the funnel, naming the segment, then showing the motion running.
Scope
Read first. You cannot instrument a site you do not own in four hours, and "who is engaging" needs historical data. So use your own site with analytics already running, or use the dataset handed out at kickoff.
Watch out
The third part is what gets scored. Instrumentation on its own is plumbing.
D3Programmatic SEOWork out what ranks for a keyword set, then build the machine that produces pages against it.
Build
Work out what ranks for a keyword set, then build the machine that produces pages against it.
Done looks like
A keyword map, a page template, and fifty pages generated live.
The judge hands you
A domain or a keyword.
Demo it by
Entering the keyword, showing the map, generating the pages on stage.
Scope
Read first. Ranking takes months, so nobody can demo the outcome today. You are judged on the machine and the pages, not on rankings.
Watch out
A growth plan document scores nothing. Ship the generator.
D4Full-Stack Content CreatorText or multimodal, end to end. Input goes in, publishable content comes out, unattended.
Build
Text or multimodal, end to end. Input goes in, publishable content comes out, unattended.
Done looks like
A finished piece you would actually publish, produced without you steering it.
The judge hands you
A topic or a repo.
Demo it by
A live run on the judge's pick.
Watch out
Fix your output format in the first twenty minutes.
A2Engager to PipelineTake a LinkedIn post with real engagement. Pull the people who engaged, qualify each against a written ICP, and output a ranked DM queue with a specific reason per person.
Build
Take a LinkedIn post with real engagement. Pull the people who engaged, qualify each against a written ICP, and output a ranked DM queue with a specific reason per person.
Done looks like
A ranked list of people, each with their verbatim comment, the URL it came from, and why they are worth a message.
The judge hands you
A post URL.
Demo it by
Showing the post, running it, then opening three rows.
Watch out
Your own posts will not have enough engagement to be interesting. Point it at a big GTM account's post instead. And do not send anything.
A3The WatcherA system that wakes up, checks ten competitor pricing, changelog and careers pages, and reports only what changed, with a diff and a link. "Nothing changed" is a valid output.
Build
A system that wakes up, checks ten competitor pricing, changelog and careers pages, and reports only what changed, with a diff and a link. "Nothing changed" is a valid output.
Done looks like
Two runs, where the second reports only the delta.
Demo it by
Comparing a run from the start of the sprint against a run from the end, with the pages that moved highlighted.
Watch out
Nothing real will change on ten pages in four hours. Seed a page you control so you have a guaranteed diff to show.
A4Sixty-Second InboundA demo request arrives with an email and a company name. Enrich it, score it, route it, and draft the first reply. Under a minute, every claim sourced.
Build
A demo request arrives with an email and a company name. Enrich it, score it, route it, and draft the first reply. Under a minute, every claim sourced.
Done looks like
A stopwatch-verifiable minute from input to a reply you would actually send.
The judge hands you
An email and a company name.
Demo it by
Starting a stopwatch on stage.
Watch out
A live run that takes four minutes instead of sixty seconds kills the room. Rehearse twice with a stopwatch and keep a pre-run version ready to cut to.