Hackathon in a VC?!
How we empowered our non-developers with ML tools
When I recently returned to work from maternity leave with my third (!) son, I wondered what challenge would ignite the professional fire in me. I’ve been working at Aleph for six years and these days I lead our Ampliphy platform team. Luckily, I returned just in time for the launch of our Ampliphy Hackathon.
Aleph invests in disruptive and meaningful innovation. When we started developing the Ampliphy platform in-house, we viewed it as an internal startup. It quickly became an integral asset of the fund that leverages our people, network, software and resources to help our portfolio companies succeed on the global stage. Ampliphy, the Aleph Value Generator, tackles the greatest startup challenges, like talent recruitment, business development and follow-on funding — at scale.
Founders and developers love hackathons — it gets everyone’s creative juices flowing and it’s also great fun. So, being at the hub of this entrepreneurial culture, we at Ampliphy run one from time to time. It breaks the routine and gives people a chance to try out crazy ideas that might not be acceptable in their “day jobs.”¹
But that’s not all a hackathon can accomplish.
Hackathon Gain #1: Developing New Skills
This time, in addition to fun, the hackathon had another objective: to encourage the non-developer members of our Ampliphy team to use machine learning (ML) in their day-to-day work by experiencing it in a casual, informal atmosphere. Ampliphy is based on advanced software and technologies; the R&D team uses ML, among other things, to create the tools that help our startups grow. However, our delivery team, as we call our non-engineering team, has rarely used machine learning.
“Delivery” is a role that we kind of invented here at Aleph. Our delivery team maps Aleph’s network assets, overlays them with automated data collection and humanizes these assets through services like talent recruiting, sales enablement and follow-on funding for our portfolio companies (and deal flow generation for Aleph’s partners). Our delivery team understood the potential of ML to enhance our working products but rarely experienced working with it firsthand. It was important to equip all team members with the capability to independently create models and train them. This would not only be a real power multiplier for Ampliphy, but would also contribute to our own professional development and enhance our productivity.
Just a few years ago, ML was a privileged technology only top engineers and developers could have used. Recently, it became more dev-mainstream, and today, with tools like Google’s AutoML, even non-developers can use it, in a relatively friendly and easy way (really!). It was clear our Ampliphy delivery team should start implementing ML as part of our routine work, and Daniel Zautner, a developer on our deal flow pod, suggested that a hackathon could be a great kickoff experiment.
Hackathon Gain #2: Innovative, Practical Tools for Our Day-to-Day Work
The goal of the hackathon was to create tools for VC operations. The projects were chosen to allow for “playing” and “familiarizing” ourselves with Google’s AutoML. Each pod focused on a project. Bar Schneidman, a graduate of business administration and psychology, trained a ML model to predict the likelihood of a newly established company to be a startup, simply based on its name and address. This ability allows the deal flow pod to identify potentially successful founders the moment they establish a new company, and sometimes even before they begin window shopping for an investor. Ampliphy obviously holds a large dataset of Israeli startups, so training the ML model for “positive” samples was relatively simple. Results were quite impressive: with ±90% precision!
Gali Baram, a financial analyst on our follow-on funding pod, trained a model to predict the likelihood of an account to be a VC based on its LinkedIn description. Venture capital firms don’t always clearly define their business in their LinkedIn name or description, making it hard to automatically differentiate them from investment banks, family offices and other investment institutions. The follow-on funding pod supports our companies in both finding and connecting them to the best investors for their company for upcoming fundraising rounds. The ability to automatically detect the venture capital entities in our network should speed up our workflows, allowing Gali to uncover opportunities that were missed until now due to incorrect classification. It was very satisfying to see the model successfully identify venture capital entities which didn’t have the words “VC” or “venture capital” in their description. Following this project, we’ve already updated the classification of hundreds of accounts in Ampliphy.
Uri Ar, Aleph’s experience and brand strategist, decided to examine, together with the help of our engineer Tal Bussel, whether it’s possible to predict if a person is a founder based on their Twitter profile blurb. They used 300 of Michael Eisenberg’s Israeli followers and then confirmed their status as founders (or not) via LinkedIn. This ability could be useful for VCs to recognize when a potential founder starts following them on social media, as well as to better understand their followers and group them according to their area of expertise and interest. It is still in the preliminary stages, but definitely allowed us to laugh a bit while we were playing with it. In the image below, you can see a screenshot of Daniel mocking the initial versions of the model, after it decided that it is 99% certain that a profile blurb stating “I am not a founder” is actually that of a, well, a founder.
We entertained various other projects as well, including training a model to predict whether a company is B2C or B2B, so we could differentiate top talent based on their experience in each field. We even created a game to collect a data set that would allow us to identify the probability that two different people from the same large company (over 500 employees) might know each other. This ability could strengthen Ampliphy’s potential to reach someone, somewhere out there in the world, who may be able to strategically help one of our startups.
Hackathon Gain #3: Personal and Professional Empowerment
We had considered embedding ML into our workflows for some time now. We always understood its potential to make our data modeling both faster and more accurate. But for us non-engineers on the Ampliphy delivery team, machine learning is just intimidating. On a personal level, I honestly doubted that I, a former lawyer, could actually create my own ML models. I was certain I would need constant help from the dev team, which would make it so much more complicated to execute (and like everyone else, I don’t like to depend on others to get my work done). But the dev team insisted this was work we could do on our own, if only we would try.
The hackathon environment turned what could have been a stressful learning experience into a fun and enjoyable experiment. We had no OKRs, no deliverables, just the goal of learning how to use ML for our workflows. Not only did we all enjoy ourselves and the opportunity to bond as a team, we actually created some useful tools we are already using, and others that we plan to continue to develop in the future.
Even more importantly, the hackathon empowered our delivery team. At Aleph, we are all owners of our domains and work independently. ML will indeed improve our workflows and help us provide better deliverables in the future without needing to depend on our engineering team. But the way in which we learned ML also boosted our own individual capabilities and power, made us feel more confident using technology in our work, and has lasting, far-reaching impact on the professional advancement and growth of each one of us.
¹ Actually the first Aleph hackathon was held for entrepreneurs and focused on tackling the issues that make the path of entrepreneurship so challenging. It was back in 2016 on Purim and included costumes, pizza and masked founders.
