Connecting People Using Software
Covalent, named for the chemical bond forged by sharing electrons, is a startup working to help organizations foster new and genuine connections between their members. With customers ranging from teams of 15 to communities of over 15,000 people, Covalent's technology needs to meet the demands of highly varied communities, cohort-based courses, and companies.
It all started when Abhishek Bhargava, Covalent's technical co-founder, reflected that "I have made a lot of amazing online friends through some of the communities I am in." Realizing the impact of nearly two years of many people's unplanned experiment with remote work, Abhishek started Covalent to help others find the same kinds of connections that he had made within his communities. Even for a repeat founder, starting a company is a lot of work, so Abhishek outsourced his infrastructure to Zeet from day one.
With Zeet I'm shipping faster, I'm getting customer feedback, and the business is working faster.
This case study examines how Abhishek has spent time that would otherwise be dedicated to infrastructure at different stages of building his startup and how he used Zeet on the side to save days of deployment on a weekend project.
Covalent isn't Abhishek's first entrepreneurial venture. Formerly, he built Vectordash, a Y Combinator-backed infrastructure startup in the GPU compute rental industry. Vectordash originally let people who owned high-end GPUs in their personal computers rent unused time to researchers and gamers while earning more than they would have made mining cryptocurrency. Later, Vectordash supplemented its marketplace by making cloud-hosted GPUs on AWS and Azure easy to rent.
Cloud gaming and a GPU compute marketplace require complex software infrastructure. As Abhishek described, "we had this big hybrid cloud solution. The peer-to-peer machines which were a pain in the ass to manage because everyone had different infrastructure from different vendors. Making sure the virtualization was robust and making sure we could actually manage all of those machines without errors was a pretty big challenge." Abhishek, a talented developer with a degree from Carnegie Mellon, was up to the technical challenge. But on his next startup, he decided to do things differently and buy instead of build as much as possible so he could focus on the customer.
Covalent's Early Days
In early 2021, Abhishek and his co-founder David identified a trend. Months before journalists coined "the Great Resignation," Covalent's founding team was thinking about how as "organizations have started to move more remote and hybrid, it has been really hard for people to feel that sense of belonging and connection that they used to pre-COVID." They quickly discovered that this lack of connection was leading to problems that companies are willing to pay to solve.
They attribute part of the recent decrease in employee retention to this reduced feeling of communal belonging. Per Abhishek, "from 2017 to 2021, the tech employee churn rate went from 13% to 20%, which is super super high. Every churned employee costs companies 50% to 200% of their salary to replace. If we can improve employee retention by 10% or 20%, that is a multi-million dollar value proposition for every company."
Identifying a real problem with a strong value proposition is a critical first step, but finding a solution is a challenge that requires everyone's input. Technical founders need to be on customer calls because no one is telling them what to build. So Abhishek split his time between development and talking to customers.
Initially, Covalent focused on discovery within communities. Abhishek first built "this glorified member directory as a service as an initial attempt to solve that problem." But, discussions with 200 community managers and people ops departments revealed that they had built a nice-to-have service that didn't address any true needs. These two months of conversations revealed major pain points in the orgs. Covalent pivoted to one such problem: fostering connections between people who are relevant to each other within organizations. Their solution, a "streamlined, robust, flexible, customizable process around connecting people who are relevant to each other in organizations," is gaining traction.
The story of this pivot rests on a classic startup moment. In a conversation with a massive potential customer one Friday, Abhishek and David "randomly ended up pitching them this connection thing, like what if we connected people who were relevant to each other at your organization." The customer excitedly asked if they could start next week, and Abhishek agreed, even though he did not have anything built out for the idea. That weekend, he wrote Covalent's matching algorithm and connected it to a Python script to send emails through the SendGrid API, introducing people and letting them know what they had in common.
Even for a repeat founder, starting a company is a lot of work, so Abhishek outsourced his infrastructure to Zeet from day one.
This weekend build left no time for configuring servers, but the opt-in flow required a live API endpoint that Abhishek had written in Django. He deployed the service with Zeet and the whole system was immediately ready for the following week. Today, Covalent is iterating on a more robust version of this weekend build and finding traction.
With over 30,000 people in their database and 10,000 matches made, Covalent is rapidly scaling, as are their infrastructure needs. What are they doing with this data? Abhishek explained, "today what we do is we connect people who are relevant to each other within organizations, and we do this in various contexts. It could be one-to-one interactions, mentorship programs, onboarding, small groups." With this data often ingested and processed in large batches from integrations like Airtable, Slack, and Typeform, Covalent's infrastructure focuses on ingesting, storing, and processing data from a variety of sources.
One distinct aspect of Covalent's architecture is its reliance on scheduled long-running tasks. "Long-running task" is a bit of a nebulous term: how long is long? As a tautological definition, you could say a long-running task is a task that runs for long enough that you have to worry about things like server restarts and data synchronization.
Covalent relies on queues and a scheduler to wrangle their long-running tasks. This infrastructure, hosted on Zeet, ensures that the processes aren't interrupted. However, if a task like data analysis takes long enough, the underlying data might change while the analysis is running. Using a job queue and task scheduler also ensures that a task runs on a synced data set.
Another high-throughput job for Covalent's infrastructure is sending communications via email, Slack, Discord, Microsoft Teams, and SMS. While they currently use SendGrid for delivery, these batches of thousands of introduction messages are heavily personalized including relevant details about the people being introduced and a Calendly link with a mutually available meeting time. Like the rest of Covalent's long-running tasks, batch email jobs pull data from PostgreSQL and are scheduled via Celery in conjunction with Redis or RabbitMQ.
Building a genuine relationship within an organization is another kind of long-running task, not a one-off action. As organizations use Covalent for years, Covalent's data sets, integrations, and processes will grow and improve to create better matches with two key goals. As Abhishek explains, "The first is foundational value alignment and the second is finding commonalities that may initially be on the surface level but over time lead to strong relationships. Over time, as people develop these relationships, they become anchored to these organizations ... This increases engagement and improves employee retention, which is a massive value prop for the organization."
As Abhishek is fostering value-aligned relationships in other communities, he is building the same within his startup and its suppliers. A regular user of Zeet's customer support, he has someone to turn to for technical support even as the sole technical person at Covalent.
I ask for help and no matter what time of day it is, someone will respond to me and my problem will be resolved in the next 30 minutes, max. That's kind of ridiculous.
Being a technical founder isn't just coding all day. Zeet gives technical founders more time for other demands like customer calls, but it also helps with tasks like onboarding new engineers and writing a roadmap.
Covalent is about to take the huge step of hiring a founding engineer, giving Abhishek someone to share his technical responsibilities with. The engineer, who lives in Switzerland, will be using Zeet from day one. As Abhishek wrote onboarding docs, he reflected, "in previous onboarding docs that I have either written or have read, there was a whole process around actually deploying code to production. In my checklist, I have stuff around getting them familiar with code and values, and the last thing on the checklist is 'Onboard to Zeet.' Those three words, it's literally that easy." As engineers join Covalent, they will enjoy the "easter egg" of not having to spend their time deploying code, instead focusing their efforts on the features they were hired to build.
Abhishek decided to buy instead of build as much as possible so he could focus on the customer.
One technology that Covalent is adding to their platform is natural language processing. These alpha-stage models will require Covalent to ingest and store unstructured data like bios and other free text. Then, the models will parse this information to find common interests and values, letting Covalent suggest more relevant connections within large organizations between people who might have otherwise never met.
This infrastructure, naturally, will be handled by Zeet. When asked about future needs, Abhishek said, "I don't worry that I will have to switch to a different DevOps provider as I grow. I'm set on Zeet and I know it will be able to handle whatever I need in the future. From a DevOps perspective, I don't need to do any planning, I just need to plan for architecture." This clear certainty makes building a roadmap a smoother process.
Side Project Support
Like most technical founders, Abhishek is a relentless builder. He uses his Zeet account for his side projects too so that he can ship his ideas quickly and get back to working on Covalent.
Zeet literally cut my time to do this whole thing by 66%, and that's absurd.
Abhishek built a successful side project in the weekend after Thanksgiving: an automation that mints an NFT whenever he gains a new follower on Twitter. The bot also DMs the NFT to the new follower, "signifying an everlasting bond" as he quipped.
Bridging web2 and web3 technologies takes some tech. The Twitter bot requires six servers: a Redis server, a Celery process, Django and React servers, a PostgreSQL database, and a Celery Beat scheduler. If working with AWS directly, Abhishek said he "would look up how to productionize each of these commands to make sure that it runs continuously and if it fails, it actually restarts on its own. That's not trivial, you have to set that up differently for Python, React, Django, whatever."
Instead, Abhishek used his Zeet account to save days of work. "I tested everything in dev, I spent 20 minutes in my Zeet dashboard setting up the project, and then everything was deployed. It was amazing. If I didn't have Zeet...I would have spent another 3 days trying to figure out how to deploy the six different servers that I need." This would have meant spending more time to deploy than the project took to develop.
When Abhishek shipped his twitter project, he got over 300 new followers in the three days that he would have otherwise spent configuring infrastructure. Instead, he could spend that time showing Covalent to a newfound audience.