From Aerospace Dreams to AI Domination Behind the Rise of Labelbox

Machine learning projects rely heavily on data. While algorithms and computing power garner much attention, the foundational reality is that artificial intelligence requires vast amounts of meticulously labeled data to function effectively. This truth was particularly evident in the early 2010s when Manu Sharma, a young aerospace engineer at Embry Riddle Aeronautical University, was experimenting with neural networks.

“At that time, working with neural networks was still archaic,” Sharma recalls. “This wasn’t even too long ago, around 2009 and 2010, but one of the best ways to work with neural networks was to use MATLAB or Simulink software packages that are widely available at educational institutions.”

What intrigued Sharma was the way these early AI systems learned. “I was very intrigued by these systems because you didn’t have to learn equations to figure them out or train them. You just fed them data and it would start mimicking behaviors it would see from patterns in the data.”

During his university years, Sharma spent a significant amount of time testing and developing neural network solutions for the aerospace industry, with a focus on autonomous flight. His classmate, Brian Rieger, shared similar interests, and the two collaborated on multiple projects before their professional paths diverged—Sharma taking product roles in Bay Area startups and Rieger joining Boeing as a data scientist in Texas.

By 2017, both were ready for a change. “We made an agreement with each other that no matter what, we were going to quit our jobs and build a company,” Sharma says. Their initial ambitions were modest—they envisioned building a small business. The challenge was determining what to focus on.

Sharma’s experience at Planet, a satellite imaging company, provided the answer. There, he observed the struggles engineering teams faced in efficiently labeling the 300-400 satellite images collected daily. “To train these neural networks on how to correctly label data—such as accurately describing what information can be pulled from a satellite image—humans have to impart intrinsic knowledge about what certain objects look like and how they can be identified,” Sharma explains.

The available tools were outdated. Teams relied on desktop software and transferred images via thumb drives. While platforms like CrowdFlower existed for basic tasks, nothing met the complexity required for satellite imagery. Sharma meticulously documented the process of building scalable AI infrastructure.”The distinct problem we set out to solve was to create a collaborative labeling system,” Sharma says. “Despite launching in 2018, all of the state-of-the-art tools at the time were desktop applications. We knew AI was going to get bigger and that data labeling would only become more important, and we wanted to build a solution that would not only manage large amounts of data but could allow multiple users on one team to log into a cloud-based platform and collaborate on the labeling process.”

Before writing code, Sharma, Rieger, and their third co-founder, Daniel Rasmuson, committed to extensive customer discovery. They developed detailed mockups and presented them to AI experts and technical leaders across industries.

“There were many experts in this field that had tried to solve the problem themselves, and they told us that Labelbox was an absurd idea that was never going to work. They said it was impossible to create a general product solution for such a bespoke problem,” Sharma recalls.

Undeterred, the team refined their feedback approach, searching for two key signals: problem validation and solution validation. “Hearing from other people that yes, they were frustrated with the desktop tools available and yes, they were working on building a solution internally and yes, they were planning to do more labeling in the future validated our assumptions on a macro level.”

A turning point came during a meeting with a medical imaging company. When the CTO attempted to hire Rasmuson to build their internal labeling tool, they knew they had struck upon something valuable. “We thought if somebody is willing to pay Dan to improve this internal process for them, that was enough signal for us to get started,” says Sharma.

They launched a beta version, posting it in machine learning forums on Reddit. “At that time, we just wanted to share what we had built with the ML community as quickly as possible and find out if it was useful or not. We took the quickest and shortest path to reach users,” Sharma explains.

The response was immediate. Within two months, publishing giant Condé Nast expressed interest, eventually signing a $20,000 contract. “When we saw Condé Nast come inbound in our second month of operating, we went to go speak with their team,” Sharma recalls. “I remember in a meeting, we signed $20,000 off the bat. Our prices started at $100 a month for our services and in the span of 60 days we were signing a $20k contract a year. We became really confident in our pricing model at that point.” Momentum built rapidly. By April 2018, they had $100,000 in annual recurring revenue. Major enterprises like Allstate and Bayer soon followed, signing six-figure contracts. By year’s end, they had 37 customers, which drew investor interest. In July 2018, they raised a $3.9 million seed round led by Kleiner Perkins and First Round Capital.

By early 2019, over a million data assets were being labeled monthly on their platform. They reached a million-dollar annual recurring revenue milestone in under a year. A $10 million Series A led by Gradient Ventures followed in June 2019, and a $25 million Series B with Andreessen Horowitz came soon after. Today, Labelbox has expanded significantly. Their platform includes a wide range of features aimed at improving AI model development. The recently launched Alignerr Connect initiative connects companies with rigorously vetted AI experts specializing in data labeling and model evaluation. This platform allows businesses to hire top-tier AI trainers, integrate them into workflows, and expand teams rapidly. This is particularly beneficial for organizations looking to enhance the efficiency and quality of their AI data pipelines.

Labelbox has also made substantial platform improvements in 2025, including a multi-modal chat editor with a built-in code runner, polygon snapping for precise annotations, and mandatory acknowledgment of project instructions. Additionally, the introduction of a labeling services marketplace enables companies to select labelers based on expertise and availability, streamlining project management.

To further improve AI model evaluation, Labelbox’s Speech Generation Leaderboard, last updated in December 2024, assesses AI-generated speech for quality, word error rate, and naturalness. Evaluations incorporate a combination of standardized metrics and human assessments conducted by Alignerrs, ensuring reliability. The multimodal chat evaluation system has also been upgraded to allow real-time interactions with multiple AI models, improving workflow efficiency and usability.

Moreover, Labelbox has integrated AI-powered code and grammar critics, acting as virtual assistants that analyze responses for errors and inconsistencies. These critics provide instant feedback, enhancing data quality at scale. A dedicated “Get Suggestions” button allows users to preview, apply, or discard suggested improvements. Through a combination of rigorous quality metrics such as Inter-rater Agreement and Krippendorff’s Alpha, and leveraging Large Language Models for quality control, Labelbox remains committed to ensuring high-quality AI training data. 

Hence the origination of the Labelbox idea: “Our biases have always been toward building tools. We’ve always loved building tools and admired great ones. So we asked, ‘What can we do here?’ Maybe we can actually come up with a product that will become an interface for these neural networks,” says Sharma.