How to hack a hackathon in 30 days? Ft. GPT-3

Published in

Cactus Tech Blog

9 min readJan 5, 2022

A lighthearted version of how to organize a virtual hackathon (🚀 the moon someday)

The word Hackathon is a portmanteau — Hack + Marathon. Hacking as a term always has had bad rep courtesy movies and pop culture across the globe. A certain other term also comes to mind (coughs in Terminator) but well for all purposes, this hacking is taken more in a positive tone. While the other term depicts a long duration, the purpose of a hackathon is to get things done, in a short burst of time. For starters, the biggest questions one needs to ask while going about hosting such events are –

1) Is it private or public?
2) What is the purpose of conducting this hackathon? Do we have clearly defined end results?
3) Evaluation Criteria
4) Prizes

I deliberately missed out on a key question that has become important since 2020 (doesn’t actually seem like we have moved past that year on the calendar). Hackathons are known for sleepless nights, endless pizza parties, and engaging activities which are only possible if you gather everyone in the same room. Yep, now those physical rooms have been replaced by Zoom/MS Teams/Video Conferencing meeting rooms.

This was an interesting challenge for me as I have been part of conducting hackathons and quizzes back when everything was as it should be, perfectly balanced with pizzas and beanbags. I had just joined CACTUS and was presented with this ask of conducting a private (employees + vendors only) virtual hackathon with the theme around the usage of GPT-3, the language model by OpenAI which shook the world last year and till very recently was in a closed beta stage and waitlist-only mode. The applications showcased even on its playground are so powerful that you can practically ideate and test out product ideas with little to no effort. All without any knowledge of ML.

Now the trick with hackathons is that there are certain types (especially tech-centric)

API-driven — You have a bunch of APIs to leverage to build something or solve given problem statement(s)
Data-driven — You are given data and a bunch of themes. You go around building everything from scratch by leveraging whatever you can find.
Platform-driven — You are made to build on certain platforms (cloud, frameworks, etc) and build keeping the theme in mind.

Running an API-driven hackathon where the API itself is tricky to get access to is not only stressful but needs a lot of planning. Things can go wrong too quickly. Coming to this part in a later section.

But, the deadline was set. In 30 days, I had to plan, co-ordinate and execute a virtual hackathon, a first in the history of CACTUS, and practically hack my way through an organization I just joined a couple of days ago and execute a bunch of ideas that would bolster the current product catalog.

As all things go, well-planned events are 50% of the battle. But what if you are just riffing ideas in one week and executing them on the fly the next? That is how it started.

September Week 1

Major Highlights (TL;dr because I can talk about it all day, some of the committee members would agree at CACTUS)

Pivot from a single event to a series of events under the house of innovation ( La casa de Innovacion)
Based on the first point, you must have guessed the theme of the hackathon i.e a popular Netflix show.
Get budgets finalized, evaluation criteria vetted and weighted.
Start with collaterals for internal marketing on slack, emails.
Prepare a kick-off presentation to explain to everyone what is GPT3 first let alone the hackathon.
Forms, forms, and forms! Registrations and Team formation.

The plus side to changing things on its head was also the fact that when you have strong leaders pushing for excellence and perfection, it tends to bring the best in whatever you are trying to do.

September Week 2

Registration week! The soft target was 50 overall registrations. The unrealistic target was 100.
Also planned a bit about going away from the traditional structure of a 24/48 hour non-stop hackathon but towards a week-long affair along with 3 days for ideation (10 days in total). As folks were also working alongside the hackathon, it also gave them a buffer period to re-iterate over the weekend.
Day 1–50 registrations.
Last day — 126 registrations — product managers, senior developers, machine learning folks, designers, business heads, everyone joined with the premise of exploring what this new kid on the block has to offer (little did they know the hell I was about to unleash in terms of meeting requests, forms, and slack messages)

September Week 3

Rejoicing at breaking through our registration goals came with a different problem — forming balanced teams.
15 teams. Spread globally across all the offices.
Kick-off!
Ideas fly, need a lot of feedback and help. Mentors to the rescue.
Mentor hours — a savior for teams to bounce ideas and get genuine feedback on feasibility, scalability, impact, and novelty.
Forms, again to sort out ties in case of similar ideas (First come, first get to execute)
50 ideas in total, 15 finally decided to be ready for build.

September Week 4

Source: https://tenor.com/WlvN.gif

BUILD.BUILD.BUILD.
Slack groups bustling with messages
API keys, IT access, GPT3 KT sessions, Product strategy sessions.
A panic attack when a team just used $700 worth of credits in an hour and exhausted the limit for the whole hackathon. As I mentioned before, the trickier the APIs, the chances of things going haywire increase so always keep buffer.
Energies start to reduce as the weekend is knocking at the door.
Pings, playlists, and motivation posts to keep everyone going.
A fun, tech banter event just before the presentation day where teams with the best memes would win a prize.
A few which are less cryptic and were really fun (according to me, so heavily biased :D)

Judgement Day

Source: https://giphy.com/gifs/awkward-pulp-fiction-john-travolta-6uGhT1O4sxpi8

A marathon 4-hour session where each team got 10 minutes to present and get cross-questioned by a panel of judges.
Most of them took more than the stipulated time but the ideas and presentations were well thought out and some of the designs were at par if not better than things in production.
Evaluation criteria kicking in. Usability, Scalability, Value, Novelty, Presentation were a few of the major factors that each team got evaluated on
We had the traditional top 3 prizes, some specific prizes which I am proud to have named :D

Eureka Moment for Best Idea
Cyberpunk Truck for Best Presentation
Mr. Robot for Most Hacky use of the API
People’s choice award for Most Enthusiastic Team

Challenges and experiments

Team Allotment

Balancing teams based on different roles looks simple but is very difficult.-
The reason why most of the hackathon organizers ask participants themselves to form a team to remove this task from their checklist.
But that results in lopsided teams comprising of only a certain skill. We took it as a challenge and tried to justify the teams as much as we could.

FAQs

Internal hackathon which runs across a week during work hours made people confused about their time commitments.
A good lesson was to compile all generic FAQs and share them with everyone which also helped in increasing registrations.

Keep the spirits high in a remote setting

Tech Banter Face-Off — People love sharing memes, tech memes being a very specific niche.
What a good way to get everyone in one place and just enjoy good, quality memes.
Some teams put effort and made memes on other teams, the event, even on me and some teams did a google search and got away with it.
All in all, I think people enjoyed this fun session which is definitely not something that would be that effective in a real-life hackathon (maybe with some live battles of roasting, taking it too far)

Evaluation and fair judgment

Having 4 hours for presentations is a lot. Equally so, making sure everything is working smoothly on a technical front is also important.
Super thankful to the Judges — Abhishek Goel, CEO, CACTUS; Nishchay Shah, CTO, CACTUS; Sourav Dutta, SVP, Strategy and Corporate Development for being patient and equally enthusiastic about the whole process.

Feedback

Helps in understanding what went well, what could be improved, and whether people would come back to attend another one
After some very professional marketing pitches and genuinely novel ideas being brought to life, it was heartening to see more than 95% of participants responding positively to attending the next hackathon and eagerly waiting for the same.

What those 30 days taught me is that practically, you can pull off a heist when you have a super supportive team and equally enthusiastic participant pool who want to be a part of a revolution —

of ideas and experiments,
of failing fast and breaking things,
of not being afraid to collaborate and working across teams and delivering at breakneck speed.

The goal was to get some good products out of the hackathon but for me, I think giving people this experience, the opportunity to present their ideas to the people at the highest level in the company was something that they will cherish which will help them grow in their current roles as the key takeaway.

If you made it this far into this journey (or just scrolled to the end, just like most of us switch to the sports section first of the newspaper), here is the shorter version of how to hack together a virtual hackathon in 30 days.

Week 1 — Plan and reach out/identify all stakeholders, finalize budget, prizes and core theme.
Week 2 — Registration marketing, setting up event collaterals in parallel, mentors, judges, blocking time on everyone’s calendar
Week 3 — Team formation, brainstorming, sessions, workshops, mentor hours, let team members get acquainted with each other.
Week 4 — Build, present, host some fun activities, evaluate (reasonably keeping your end goal for the hackathon in mind) and prize distribution.

If all goes well, at the end of it, you will be writing a report on how everything happened, presenting it at various company meetings, and still writing about it after 4 months.

One of the ideas revolved around automatically generating minutes of the meetings and using them to act as a knowledge base so that whenever you would forget what happened in the meeting, you could just ask the bot. I used it to write this blog as even I had forgotten things that had happened.

Just kidding, I just saw their presentation as an inspiration for it was so powerful that it would bring Steve Jobs back from the afterlife and use their pitch to present the new iPhones.

Oh, and if you do want to be part of a team of tinkerers and dreamers, we are always hiring!