Radha Basu, Founder and CEO of iMerit has built her career at HP, spending 20 years with the tech giant and eventually heading its Enterprise Solutions group. She then took Support.com public as its CEO. Radha started Anudip Foundation in 2007 with Dipak Basu and then founded iMerit in 2012. She is considered a leading tech entrepreneur and mentor, and a pioneer in the software business.
iMerit delivers multimodal AI data solutions by combining automation, expert human annotation, and advanced analytics to support high-quality data labeling and model fine-tuning at scale.
You’ve had a remarkable journey—from building HP’s operations in India to founding iMerit with a mission to uplift marginalized youth in Bhutan, India, and New Orleans. What inspired you to start iMerit, and what challenges did you face in creating an inclusive, global workforce from the ground up?
Before founding iMerit, I was Chairman and CEO of SupportSoft, where I led the company through its initial and secondary public offerings, establishing it as a global leader in support automation software. That experience showed me the power of combining people and technology from day one.
While India’s tech boom created new opportunities, I noticed many talented young people in underserved areas were left behind. I believed in their potential and drive to learn. Once they saw how software could power advanced technologies like AI, they eagerly embraced these careers.
We launched iMerit with a small, diverse team, half of whom are women, and have grown rapidly ever since. Our team’s adaptability and coachability have been key, especially as data-centric AI has increased long-term demand for skilled specialists.
Today, iMerit is a global provider of AI data solutions for mission-critical sectors like autonomous vehicles, medical AI, and technology. Our work ensures customers’ AI models are built on high-quality, reliable data, which is essential in high-stakes environments.
Ultimately, our strength lies in strong technology underpinnings and a team of well-trained, motivated employees who thrive in a supportive, learning-driven culture. This approach has fueled our growth, kept us cash positive, and earned us high NPS scores and loyal clients.
iMerit now works with over 200 clients, including tech giants like eBay and Johnson & Johnson. Can you walk us through the company’s growth journey—from those early days to becoming a global leader in AI data services?
We’ve had a front-row seat to our clients’ AI journeys, partnering from early experiments to large-scale production. Our work spans startups, global autonomous vehicle leaders, and major enterprises. By training their models from the ground up, we’ve gained unparalleled insight into what it truly takes to scale AI in the real world.
The field has evolved constantly and rapidly. I have rarely seen a technology advance so dramatically in such a short time. We’ve transformed from a data annotation provider into a full-stack AI data company, delivering specialized solutions across the entire human-in-the-loop (HITL) lifecycle: annotation, validation, audit, and red-teaming. Handling edge cases and exceptions is vital for real-world deployment, requiring deep expertise and nuanced judgment at every step.
Our largest vertical is autonomous mobility, where we manage the full perception stack, including sensor fusion across 15 sensors for passenger, delivery, trucking, and agricultural vehicles. In healthcare, we drive clinical imaging AI. In high-tech, we’re at the forefront of GenAI tuning and validation, demanding greater sophistication in our workflows and talent.
Success in these domains isn’t just about having experts- it’s about cultivating expertise: the cognitive ability to challenge, coach, and contextualize AI models. This is what sets our teams apart.
Our growth is fueled by long-term partnerships, and most of our top ten clients have been with us for over five years. As their needs grow more complex, we continually elevate our domain knowledge, tooling, training, and solutions. Both our tech stack and our people must constantly evolve.
The fusion of software, automation, annotation, and analytics, creates the rubric for very flexible, rapid, highly precise, human-in-the-loop interventions. 70% of new logos are on our own tech stack, which requires a huge internal transformation. Again, our culture ensures the teams are hungry to learn and want to grow constantly.
What have been the most pivotal moments in iMerit’s history—whether technological milestones or strategic decisions—that helped shape the company’s trajectory?
At a time when AI data work was seen as crowd-based gig work, we took an early bet that this would grow as a career and would require complexity and enterprise focus. By building in-house teams dedicated to advanced use cases, we enabled our clients to scale rapidly, culminating in our first $1M MRR deal in autonomous vehicles, a milestone that validated our approach.
The COVID-19 lockdown tested our agility: we transitioned from fully in-office to fully remote almost overnight, investing heavily in infrastructure, security, and culture. Within weeks, client operations rebounded, and we grew both revenue and headcount that year. Today, with 70% of our team back on-site, we continue to leverage remote talent, launching Scholars, our global network of subject matter experts for GenAI tuning and validation. Whether it’s a cardiologist or a Spanish mathematician, our high-touch culture attracts and motivates top talent, directly elevating the quality and consistency of our solutions.
In 2023, we acquired Ango.ai, an AI-powered data labeling and workflow automation platform, to drive the next generation of AI data tools. This pivotal move merged iMerit’s domain expertise with Ango’s advanced tooling, expanding our capabilities in radiology, sensor fusion, and GenAI fine-tuning. We still work with customer tools as well, but many new clients are now onboarded directly to Ango Hub, drawn by its user-friendly workflows and robust security, which are essential requirements in our industry.
Enterprises consistently tell us they’re looking for the best of both worlds: expert human insight to ensure quality, combined with a secure, scalable platform that delivers automation and analytics. Combining forces with Ango delivers exactly that, uniquely positioning us to meet the complex demands of today’s most ambitious AI projects and scale with confidence.
iMerit is deeply involved in advanced domains like autonomous vehicles, medical AI, and GenAI. What are some of the unique data challenges you face in these sectors, and how do you address them?
Data-related tasks typically account for nearly 80% of the time spent on AI projects, making them a critical component of the pipeline. The data-centric part of AI can be time-consuming and expensive if not handled appropriately and scalably.
Data quality, and especially the avoidance of egregious errors, is essential in mission critical sectors that we operate in. Whether it’s a perception algorithm or a tumor detector, clean data is essential in the training-to-validation loop.
Exception handling is disproportionately valuable. Human insight into why something is outside the norm or why a scenario broke the model creates massive value in making the model more complete and robust.
In addition, context windows are becoming larger. We’re summarizing clinical notes of an entire doctor-patient consultation and analyzing anomalies in MRIs based not only on the image but also on the patient’s medical context. Subject matter experts have to set up rubrics to analyze the data accurately and ensure quality.
Safety, privacy, and confidentiality are hot button topics. Our Chief Security Officer has to safeguard against unauthorized access, deletion, and storage of data. Infosec protocols like SOC2, HIPAA and TISAX, have been major areas of investment for us.
Finally, our engineers and solution architects are constantly working on custom integrations and reports so that unique customer needs are reflected in the last mile. A one-size-fits-all approach doesn’t work in AI.
You’ve spoken about combining robotics and human intelligence as a safer path for AI. Can you expand on what that workflow looks like in practice—and why you believe it’s better than trying to eliminate AI’s creative divergence?
AI provides scale, meaning that companies are developing tools to automate lengthy processes traditionally carried out by humans. But humans provide the last mile of flexibility, certainty and resilience. As software-delivered services continue to proliferate in AI, the most successful companies will effectively combine robotics with Human-in-the-Loop practices (HITL).
We see HITL as a consistent layer in every phase of the AI development and deployment lifecycle, and also as a pillar of trust and safety. Consequently, human intelligence will be essential to course correct if the models fail. These critical applications will need the human mind to determine what changes will need to be made. This is where HITL services will become even more significant as we integrate AI into production and field operations.
Your Ango Hub platform blends automation with human-in-the-loop expertise. How does this hybrid model improve data quality and model performance in production AI systems?
AI and automation provide scale and speed, while humans provide nuance, insight and oversight. HITL ensures human involvement at critical junctures in the AI lifecycle – ensuring high-quality inputs, validating outputs, identifying edge cases, fine-tuning models for domains, and providing contextual judgment. Humans help ensure accuracy by reviewing and verifying outputs, catching hallucinations or logic errors before they cause harm. They also provide oversight in ethically sensitive or high-risk contexts where LLMs shouldn’t make final calls. More importantly, human feedback fuels continuous learning, helping AI systems align more closely with user goals over time.
HITL takes many forms. Human experts engage in targeted annotation, apply complex reasoning to edge cases, and review AI-generated content using structured QA interfaces. Rather than evaluating every decision, contextual escalation systems are often implemented. These systems route only low-confidence outputs or flagged anomalies to human reviewers, balancing oversight with efficiency.
Another critical use of HITL is fine-tuning AI agents via Reinforcement Learning from Human Feedback (RLHF). Human reviewers rank, rewrite, or provide feedback on agent responses, which is especially important in sensitive domains like healthcare, legal services, or customer support. In tandem, scenario-based testing and red teaming allow human evaluators to test agents under adversarial or unusual conditions to identify and patch vulnerabilities pre-deployment.
AI’s full potential is realized only when humans remain in the loop, guiding, validating, and improving each step. Whether it’s refining agent outputs, training evaluation loops, or curating reliable data pipelines, human oversight adds the structure and accountability AI needs to be trusted and effective.
With Generative AI tools evolving rapidly, how is iMerit staying ahead in providing evaluation, RLHF, and fine-tuning services?
We recently launched the Ango Hub Deep Reasoning Lab (DRL), a unified platform for Generative AI tuning and interactive development of chain-of-thought reasoning with AI teachers. Our DRL enables real-time, turn-by-turn processes and evaluation based on human preferences, leading to more coherent and accurate model responses to complex problems.
Advances in GenAI models and application development highlight the value of clean, expert-created, validated data. With the Ango Hub DRL, experts can test models, identify weaknesses, and generate clean data using chain-of-thought reasoning. They interact with the models in real-time and send prompts and corrections back step-by-step in a single interface.
Leveraging iMerit Scholars, the Ango Hub DRL refines model reasoning processes. It leverages iMerit’s extensive experience with HITL workflows. Experts design multi-step scenarios for complex tasks, such as creating chain-of-thought prompts for advanced math problems. iMerit Scholars review outputs, correct errors, and capture interactions seamlessly. The magic is not in onboarding large numbers indiscriminately. The best Mathematicians aren’t necessarily the best teachers. One also shouldn’t treat a cardiologist like a gig worker. The fitment and coaching of subject experts to think in the ways that benefit the model training process the most, as well as the engagement, make the difference.
What does “expert-in-the-loop” mean in the context of fine-tuning generative AI? Can you share examples where this human expertise significantly improved model outputs?
Expert-in-the-Loop combines human intelligence with robotic intelligence to advance AI into production. It involves human experts who validate, refine, and enhance the outputs of automated systems.
Specifically, expert-led data annotation ensures that training data is accurately labeled with domain-specific knowledge, thereby improving the precision and reliability of predictive AI models. By reducing biases and misclassifications, expert-driven annotation enhances the model’s ability to generalize effectively across real-world scenarios. This results in AI systems that are more trustworthy, interpretable, and aligned with industry-specific needs.
For example, after acquiring a large corpus of medical data, an American multinational technology company needed to evaluate the data for use in its consumer-facing medical chatbot to ensure safe and accurate medical advice for users. Turning to iMerit, they leveraged our extensive network of US-based healthcare experts and assembled a team of nurses to work in a consensus workflow with escalations and arbitration provided by a US Board Certified physician. The nurses began by evaluating the knowledge base featuring definitions to assess accuracy and risk.
Through edge case discussion and guideline revision, the nurses could reach consensus in 99% of cases. This allowed the team to revise the project design to a single-vote structure with a 10% audit, thereby reducing project costs by over 72%. Working with iMerit has enabled this company to continually identify ways to scale medical data annotation ethically and efficiently.
With over 8,000 full-time experts worldwide, how do you maintain quality, performance, and employee development at scale?
The definition of quality is always tailored to each client’s specific use case. Our teams collaborate closely with clients to define and calibrate quality standards, employing custom processes that ensure every annotation is rapidly validated by subject matter experts. Consistency is important to the development of high-quality AI. This is supported by high employee retention (90%) and a strong focus on production analytics, a key differentiator in the design of Ango Hub, shaped by daily user input from our team.
We continually invest in automation, optimization, and knowledge management, underpinned by our proprietary iMerit One training platform. This commitment to learning and development not only drives operational excellence but also supports long-term career progression for our employees, fostering a culture of expertise and growth.
What advice would you give to aspiring AI entrepreneurs who want to build something meaningful—both in technology and in social impact?
AI is moving dizzyingly fast. Go beyond the tech stack and listen to your customers to understand what matters to their business. Understand their appetite for speed, change and risk. Early customers can try things out. Bigger customers need to know that you are here to stay and that you will continue to prioritize them. Set them at ease with your proactive approach towards transparency, safety and accountability.
Additionally, carefully select your investors and board members to ensure alignment on shared values and concerns. At iMerit, we experienced significant support from our board and investors during challenging times such as COVID-19, which we credit to this alignment.
The key qualities that contribute to an entrepreneur’s success in the tech industry go beyond taking risks; they involve building a profitable, inclusive company.
Thank you for the great interview, readers who wish to learn more should visit iMerit.
Credit: Source link
