Everyone evaluates the opportunity. The builders who level up fastest also ask: Who will this put me around?

That's been Yusuf Sallam's instinct since he was in high school. He sought out rooms filled with people who were smarter, more experienced, and further ahead than he was. Over time, those rooms shifted. He had, without quite planning it, become someone others sought out in return.

Also in this issue: 

Read time: 7 minutes

When presented with a work opportunity, most people look at what they’ll be doing and the pay.

The people who grow fastest also look at who they'll be building with. They ask: Who else is in the room?

Across Yusuf Sallam’s journey from school coding clubs to hackathons and startups, that’s been the guiding question. Is this a room where everyone is smarter? Is this a community where people are building things I don’t yet understand? Is this a company where the standard is higher than anywhere else I’ve been?

Last week, Brilliant's AI tutor Koji launched and went viral. It was also Yusuf's first week at the Reach portfolio company.

When we talked, it was his third day. He was grinning, and carrying the kind of excitement that makes other people want to be in the room. These are the rooms that he’s been chasing his whole life.

***

As a freshman, Yusuf joined his high school cybersecurity club. Everyone was expected to contribute to the club's annual competition, Bergen County Academies Capture The Flag (BCACTF).

He spent two weeks trying to fix a bug. An older student fixed it in five minutes. As humbling as it was, what Yusuf took away was: “I need to be around people like this more often.”

“I always made it a point to surround myself with people who knew significantly more than me,” Yusuf told Reach. “Because in those environments, where I felt uncomfortable and had no idea how to contribute, those were the environments where I learned the most.”

The freshman writing buggy code eventually helped grow BCACTF into an international competition with thousands of participants. When he didn’t like the software the competition ran on, he decided to rebuild the platform himself.

He dramatically underestimated the amount of work involved, but that was a silver lining.

“I was just kind of too stupid to think it was a bad idea, so I kind of just did it anyways,” he said. “A healthy amount of naivety is pretty convenient.”

***

When Yusuf arrived at Rutgers, the builder culture he loved in high school wasn't there. So he and his roommate Safa went looking for it.

Hackathon after hackathon.

That's where he first emerged on our radars, at a Major League Hackathon event. Their hunger and drive to build resonated with our team and with our founders.

Before their first college winter break, TeachShare called. The fast-growing startup was looking for engineers to build an ML pipeline that analyzes education documents to help teachers create personalized instructional materials faster.

But there was one problem: neither Yusuf nor Safa knew machine learning.

“We were over the moon,” Yusuf said. “We could not blow this.”

For three weeks, they spent winter break teaching themselves everything they could.

No usable dataset existed online, so they built their own. They annotated examples by hand, one by one, just to get the model off the ground. At first, the dataset was too small for fine-tuning to work well. Once they realized the bottleneck was data, they focused on generating more of it. Their workflow became recursive: train a model, use it to help annotate additional examples, then train again.

Eventually they started scaling the datasets. But more data required more compute.

Which led to another problem: No GPU access.

But Yusuf remembered Rutgers was a research institution. Students could request batch compute resources.

"Safa would request four GPUs on his account. I'd request more on mine. Every night, we'd make sure the training jobs were still running before going to sleep. Every morning, we'd wake up and check what came back." 

The project worked. The model performed better than expected, and they finished TWO weeks earlier than expected.

***

After this trial by fire, doors opened. Through TeachShare founder, he was introduced to Z Fellows, a one-week bootcamp and accelerator program designed for early-stage builders with mentors such as the founders of Netflix, Angellist, Tinder, Ramp, and more.

Inspired by that community of exceptional builders, he co-founded the Venture Capital Club at Rutgers. In just a year, the community has partnered with Reach, Kleiner Perins, a16z, First Round and others to demystify early-stage investing, learn about sourcing startups, and connect with mentors and builders.

And by sharing his profile and projects on The Repo Network, he was spotted by Jared Silver, who leads the engineering team at Brilliant. After reading the young builder’s profile, Jared reached out to talk. Then he offered Yusuf a Summer internship.

Discomfort is a throughline in Yusuf’s story, but he’s hardly alone. AI models and tools are moving fast enough to make even experienced builders feel daunted.

But at fast-growing startups, qualifications don’t come from credentials. They come from learning and working quickly — trying, making mistakes, and improving. And seeking to be in those rooms where people raise your standards. They come from asking questions. 

“Worst that can happen is you hear a no," Yusuf said. "Best case scenario is you get what you want."

Building paths for students and early talent access meaningful career opportunities.

Developing the intelligence layer for clear, confident hiring at scale.

Can a thirty-second voice memo catch a disease a blood test would miss?

The instinct is to be skeptical. Real diagnostics happen with needles, scans, and labs, not a microphone.

But that skepticism misses what the research already shows. The way you speak carries objective, measurable signals about respiration, neurology, and brain control, all at once. Speech changes can precede Parkinson's motor symptoms by up to a decade. A vocal jitter tracks with depression. One study identified diabetes from voice alone in roughly 70% of cases. 

Voice is the rare biomarker that's non-invasive, multisystemic, and constant. You can collect it without a needle or a clinic, hundreds of times a day, from a phone someone already owns. The signal was always there. We just didn't have the tools to read it… yet.

Reinvent the recovery score, minus the wearable: Most recovery scores read your body through a device you have to charge and remember to wear, like a Whoop or an Oura ring. Voice could surface different signals about stress, cognitive load, respiratory effort, and mood. Build the device-free alternative to Whoop, or the layer that plugs into Oura and Garmin and supercharges the scores they already give.

Catch the decline between appointments: Chronic conditions like heart failure, COPD, and depression flare in between doctor’s visits, when no one is measuring. A daily voice check-in can catch that flare early and help clinicians check in if needed. Speech changes can surface in Parkinson's patients a decade before tremors do. The strongest evidence so far is in heart failure, COPD/asthma, depression/anxiety, and neurodegenerative diseases, so start there. Novoic is doing this for early Alzheimer's, Cordio for congestive heart failure, and Ellipsis Health for mental health. 

Build the layer everyone else plugs into: Every voice health product needs the same plumbing underneath: clean audio pipelines, multilingual models, HIPAA-grade privacy rails, and APIs other health apps can wire into in an afternoon. This is the Twilio or Stripe of voice health, the unglamorous layer everything downstream depends on. Some early companies in this category include Kintsugi, Sonde Health, and Canary Speech, and there’s yet to be a standard.

Our full thesis: http://reachcapital.com/resources/thought-leadership/voice-digital-health-sensor/ 

Happy Summer,

Valentina and the Reach team

Reach Capital invests in early-stage founders redefining how we learn, live, and work. Our portfolio of 130+ startups includes tools you might’ve used in school (ClassDojo, Desmos, Brilliant) and next-gen, AI-native disruptors shaping how future generations build, work, and thrive (like Replit and GPTZero). 

These teams are constantly on the lookout for talented builders. Share what you’re studying, building, or exploring, and we’ll intro you to projects, people, and paid opportunities for you to build upon.