iMerit Technology

Public benchmarks feel reliable until scores start clustering near the ceiling and stop separating strong models from ones that simply learned the test. That's benchmark saturation, and it happens to every static, public evaluation set once frontier model testing catches up to it. iMerit builds custom evaluation sets based on customer data and requirements, with domain experts embedded across math, healthcare, and manufacturing through its Scholars network and Ango Deep Reasoning Lab. Continuous refresh cycles and expert-calibrated scoring keep these domain-specific AI benchmarks discriminative long after public benchmarks have topped out, turning evaluation into an ongoing AI evaluation pipeline rather than a one-time deliverable. Read the blog: https://lnkd.in/g9x8-Dpc  #FrontierAI #AIEvaluation #BenchmarkSaturation #AIBenchmarks #MachineLearning

Carbon Robotics has replaced its crop-specific AI vision models with a single large plant model, and iMerit delivered the training data and workflow behind it. The results behind that shift: ▪️1M+ plant images annotated ▪️184+ plant types identified ▪️61% reduction in time to complete the most complex scenes ▪️Up to 500,000 weeds eliminated per hour The new model lets farmers customize laser weeding instantly. Tag a few images as "crop" or "weed" on an iPad, and the model adjusts on the spot, no retraining, no new software. The same model can zap weeds in an Arizona carrot field, then reconfigure for lettuce on another farm in minutes. Carbon Robotics CTO Alexander Sergeev told The Robot Report that the model needed to compare new field examples against what farmers had already tagged as crop or weed, and that comparison had to happen fast. Read the case study: https://lnkd.in/gjtHfWHn Read the coverage: https://lnkd.in/gTsZxUaj #PhysicalAI #AgTech #AIinAgriculture #FoundationModels #Robotics

Anwesa Choudhuri, Senior Research Scientist at United Imaging Intelligence, spoke with iMerit's Vimal Patel about what it took to build MedVidBench, a benchmark spanning laparoscopic, open, and robotic surgery, plus nursing. She breaks down the hardest parts of getting ground truth right across four very different clinical domains, why tasks like Critical View of Safety assessment demand real clinical judgment rather than just a labeling guideline, and what typically goes wrong when medical video benchmarks chase leaderboard rankings instead of rigor. At iMerit, our Medical pod works on exactly this kind of problem daily, annotating complex clinical videos for phases, actions, and critical structures across surgical and clinical settings. Conversations like this connect the benchmark side with the labeling side directly. MedVidBench is featured as part of the MedVidU workshop challenge at ECCV 2026. Catch team iMerit at Booth #22, September 8–12, in Malmö, Sweden. There's more to the conversation. Watch the video. #MedicalAI #AIinHealthcare #SurgicalAI #DataAnnotation #HealthTech #ECCV2026 European Conference on Computer Vision

Most 3D point cloud failures don't show up in the data itself. They start at the sensor, carry through calibration, and compound during annotation, long before a model is ever evaluated. A pedestrian twenty meters out might be just a handful of LiDAR points. A miscalibrated sensor can shift grasp estimation on a robotic arm or throw off object placement at highway speed. Ango Hub's 3D Point Cloud Tool catches these issues directly, through temporal frame traversal, cross modal validation, and stage by stage QA. Read the blog to learn more: https://lnkd.in/gPHFJY8z #Robotics #AutonomousVehicles #ComputerVision #LiDAR

Farmer voice queries come with dialects, crop and disease terminology, and noisy real world audio that most speech models were never trained on. Building ground truth data for that meant going beyond generic transcription. iMerit partnered with Digital Green on a human in the loop pipeline built on Ango Hub: ▪️Native language experts transcribed real farmer voice queries ▪️Multi layer QA turned raw audio into structured, benchmarkable data The resulting dataset is helping pinpoint where speech models misfire on regional terms and fine tune toward more accurate, trustworthy recommendations at scale. Read the case study: https://lnkd.in/giX83ctM #VoiceAI #AgricultureAI #PrecisionAgriculture #DataAnnotation #AIforGood

Text, image, audio, and video each demand different things from a model. Spatial boundaries for images, cultural and linguistic context for text, precise timestamps for speech, and consistent object identities across frames for video. Treating all four the same way produces weak ground truth and models that fail in the real world. iMerit builds a distinct workflow for each modality: ▪️Text: subject matter experts for nuanced, domain heavy labeling ▪️Image and video: AI assisted pre labeling with multi stage QA ▪️Audio: native language annotators for accents, dialects, and noisy recordings Read the blog: https://lnkd.in/gakWHvCk  #AITraining #MachineLearning #ComputerVision #AutonomousVehicles

𝗪𝗵𝘆 𝗘𝗫𝗟 𝗦𝗽𝗲𝗻𝘁 $𝟯𝟭𝟬 𝗠𝗻 𝘁𝗼 𝗠𝗮𝘀𝘁𝗲𝗿 𝘁𝗵𝗲 𝗘𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗔𝗜 𝗗𝗮𝘁𝗮 𝗘𝗻𝗴𝗶𝗻𝗲? iMerit is now part of EXL. As part of the deal, Radha Basu, Founder and CEO of iMerit, joins EXL as Executive Vice President and Head of iMerit and now sits on its executive committee. In an editorial feature by Siddharth Jindal, Senior Tech Journalist at AIM, Rohit Kapoor, Chairman and CEO of EXL, and Radha Basu share their insights on moving enterprise AI from experimentation to reliable, production-scale outcomes, with healthcare, banking, and physical AI as key proving grounds. Read the full story: https://lnkd.in/gQ5t2pyA #EnterpriseAI #PhysicalAI #Robotics #AutonomousVehicles #AI Vivek Jetley, Vikas Bhalla, Bhupender Singh, Vishal Chhibbar, Anand Logani, Narasimha Kini, Anita Mahon, Maurizio Nicolelli, Ajay Ayyappan, Baljinder Singh, Pam Harrison, Phillip (PJ). DD Ganguly, Tapishnu Chaudhuri, Anupam Biswas, Sudeep George, VG Shri Rajini Prriya, Jai Natarajan, Kerri Reynolds, Benjamin Wood, Vishnu Vardhan Nadella, Brett Hallinan.