How Fellow built its AI meeting recorder with Recall.ai

Executive Summary

Challenge

Fellow wanted to add meeting recording and transcription to its AI meeting productivity platform, but building the infrastructure internally would have required an estimated three months of work from three full-time engineers and an SRE.

The team also needed that infrastructure to be highly reliable, support multiple meeting platforms, and handle ongoing maintenance and speaker diarization.

Solution

Fellow built its meeting recording functionality on Recall.ai.

Recall.ai captures meetings for Fellow across Zoom, Google Meet, and Microsoft Teams and provides the meeting data Fellow needs, including recordings, transcripts, and speaker information.

Instead of building and maintaining meeting recording infrastructure itself, Fellow could focus its engineers on turning meeting data into useful features for its customers.

Results

With Recall.ai, Fellow:

• Launched meeting recording in approximately one week instead of its estimated three months
• Avoided dedicating three engineers and an SRE to building its own meeting recorder
• Records and transcribes scheduled and ad hoc meetings across Zoom, Google Meet, and Microsoft Teams
• Uses Recall.ai’s speaker data to identify who said what without building its own diarization pipeline
• Avoids the ongoing maintenance and on-call requirements of operating meeting recording infrastructure
• Keeps its engineering team focused on Fellow’s core AI meeting product

Fellow needed meeting recording without building the infrastructure itself

Fellow is an AI meeting productivity platform designed to help teams get more value from their meetings through collaborative agendas, action items, AI-generated summaries, and insights.

To expand what Fellow could do with meeting data, the team wanted a way to capture audio and video from meetings, save and transcribe recordings, and identify individual speakers.

Initially, Fellow considered building the entire system itself.

“We were just starting to explore the possibility of having a bot that joins meetings to capture the data we wanted — but we weren’t sure where to start,” said Alexandra Sutherland, Director of Engineering at Fellow.

The team estimated that building its own meeting recorder would take approximately three months of intensive work.

“The project was going to require three full-time engineers and an SRE to build, so it would’ve taken a lot of time away from building features for the rest of the product,” Sutherland said.

And building it was only the beginning. Meeting recording would become critical infrastructure for Fellow, meaning the team would also need to keep it running reliably.

“There’s a lot that goes into maintaining something like this, and it’s extremely important that there’s no downtime,” Sutherland said. “So we were trying to figure out if we need a 24/7 on-call person, ready-to-go at a moment’s notice.”

For Fellow, that raised a larger question: was meeting recording infrastructure really where its engineers should be spending their time?

“Our customers care about the value of what we’re providing them,” said Patrick Gingras, senior software engineer at Fellow. “To provide value, we need the meetings recorded, but we really need to spend our time developing the core features of the product — not the implementation details.”

Recall.ai gave Fellow meeting recording across Zoom, Google Meet, and Microsoft Teams

Instead of building the meeting recording layer internally, Fellow integrated Recall.ai.

Today, Fellow uses Recall.ai to record, save, and transcribe scheduled and ad hoc meetings across Zoom, Google Meet, and Microsoft Teams.

The implementation was significantly faster than Fellow’s original plan.

The team had estimated that its internal recording system would take approximately three months to build. With Recall.ai, Fellow had meeting recording working within a week.

“I felt very relieved from the second I saw the first prototype of Recall,” Sutherland said. “I had in the back of my mind all the things that could go wrong, but within one demo I saw how happy the engineers were knowing they weren’t having to maintain a complex system. The progress we made was so fast.”

That let Fellow move on from meeting infrastructure and spend its engineering time on the functionality its customers interact with directly.

Recall.ai handles speaker diarization for Fellow

Capturing the meeting was only part of the problem.

For Fellow to generate useful meeting insights, it also needed to understand who said what. Building that speaker-identification pipeline would have added another layer of complexity to Fellow’s infrastructure.

Recall.ai provides Fellow with speaker information alongside the transcript.

“Recall takes away the pain of diarization completely,” Sutherland said. “It matches up the speakers to their names, transforms the transcript with the data it has about speakers, and sends that to us directly so we don’t need to worry about any of the diarization process.”

That gives Fellow structured meeting data it can use to build its own AI-powered meeting experiences without maintaining the underlying diarization infrastructure.

Reliability lets Fellow treat Recall.ai as critical infrastructure

Outsourcing a core piece of Fellow’s meeting product required the team to trust that infrastructure in production.

Reliability was therefore one of Fellow’s primary considerations when evaluating Recall.ai.

“Reliability was one of our top priorities and, when we plugged Recall in, it just worked,” Gingras said. “It did what we needed it to do and it let us move on to working on our other features.”

That reliability also removes a maintenance burden Fellow had anticipated if it built meeting recording itself, including potentially staffing around-the-clock support for its recording infrastructure.

Recall.ai’s team has continued to work with Fellow as its requirements have evolved.

“Their team is extremely open to hearing new ideas, feedback, and thoughts,” Sutherland said. “They’re always willing to collaborate together and come up with new solutions to fit our use case.”

Fellow launched meeting recording in one week instead of three months

The most immediate impact for Fellow was speed.

Its initial estimate called for three engineers and an SRE working for approximately three months to build the meeting recording system.

Recall.ai allowed Fellow to get the functionality working within approximately one week.

“If I had to guess, we might’ve had four people for several months doing what Recall does for us in a couple of weeks,” Gingras said.

But the longer-term benefit is that Fellow doesn't have to keep allocating those engineers to the recording layer as the product grows.

“It’s powerful how we don’t have to worry about the recording aspect of things,” Gingras said. “Ultimately, it’s allowed us to bring more value to our customers, because we can focus on solving the problem we originally set out to solve: leveraging the knowledge that gets trapped in meetings.”

Sutherland put the benefit more simply:

“Recall will help you get your product into the hands of customers much faster. It’s a terrific product that I absolutely recommend.”

By using Recall.ai for meeting recording, transcription, and speaker data, Fellow can keep its engineering team focused on the layer that differentiates its product: helping customers turn conversations into useful knowledge.