Hyderabad, India
Founder, ClutchLabs AI
I’m a founder and AI engineer with seven years of experience turning research into products and commercial outcomes. My work spans diagnostic systems that reached international markets, enterprise automation that generated millions in revenue, and frontier-model evaluation. At ClutchLabs, I’m building physical intelligence that learns from human experience.
Computer vision across medical imaging, documents, and video. Connecting model development to products that operate at scale.
End-to-end AI systems: problem definition, modeling, deployment, and optimization, tied to measurable commercial outcomes.
Clinical validation, regulatory evidence, and frontier-model evaluation. Establishing the evidence needed to deploy with confidence.
Products built. Revenue unlocked. Markets opened.
I founded ClutchLabs to make human experience usable by machines. The ambition is to turn recorded human activity into a foundation for robots that understand tasks and operate in the physical world.
The pipeline starts with perception: recovering hand motion, geometry, and physical state from video. Reconstruction and quality evaluation establish what can be trusted; transferring that experience into robot learning is the next stage. My focus is on connecting these stages into a dependable system, with commercial licensing and measurable performance built into the engineering decisions.
ClutchLabs combines published research infrastructure with experience shipping a live vision product. ClutchRecon serves two national-team coaching partners, demonstrating video-to-structured-data capability in a commercial setting. ClutchLabs also has support from The Founding Co., The Residency, NVIDIA Inception, and AWS Startups.
I published EgoBench to evaluate hand-pose reconstruction across 112 tasks and 530 clips. It provides reproducible comparisons, per-task error analysis, and licensing information to guide reconstruction decisions.
EgoBench evaluates calibrated EgoDex recordings. Robot-learning transfer is an ongoing development objective; ClutchRecon traction relates to the esports SaaS.
ClutchRecon turns match recordings into tactical intelligence for esports teams. Coaches can move from a full VOD to rounds, player metrics, and the underlying events, with video evidence attached to the analysis.
I developed the vision pipeline and brought it into a live SaaS product. Detection, minimap segmentation, and projective geometry reconstruct game state directly from Valorant footage. The product connects that information to a synchronized analysis workspace without relying on a game API.
Two national-team coaching partners. The live product has analyzed 269 matches, tracking 43,979 kills across 5,965 rounds.
For coaches, the result is a review workflow built around searchable, seekable match evidence and player-level analysis.
Usage figures published by ClutchRecon, September 2026.
At Foundation AI, I worked on automating knowledge extraction and mapping across thousands of legal and insurance documents each day. The system connected information across case hearings, depositions, and insurance records, including workers’ compensation matters.
My work spanned document vision, knowledge extraction, and mapping information across related records. I also developed an alternative modeling approach for a production system whose performance put major client relationships at risk, bringing it above the client’s acceptance threshold.
Approximately $2M in additional annual revenue enabled by document automation and knowledge extraction.
Approximately $800K in annual revenue retained by improving performance for clients at risk of churning.
The outcome was both expanded commercial value and stronger retention: automated processing at enterprise volume, with performance that clients could continue to rely on.
At SigTuple, my work combined medical computer vision with the clinical-study automation needed to validate diagnostic devices. This included urine sediment analysis, ground-truth methodology, and model error characterization.
Clinical-study workflow cycle time reduced from three weeks to 30 minutes. Automating the workflow accelerated the generation of clinical validation evidence and removed a major operational bottleneck.
The urine-sediment vision models achieved greater than 96% sensitivity and specificity, providing the performance evidence needed for regulatory review.
The modeling and validation work supported FDA clearance and IVDR certification, enabling international commercialization of the devices.
Founded ClutchLabs to build physical intelligence from human experience. Published EgoBench and shipped ClutchRecon, a live SaaS serving two national-team coaching partners.
Delivered LLM evaluation and tool-use task work for Anthropic’s frontier-model program through Turing’s external contractual team.
Automated knowledge extraction at thousands of documents per day, enabling ~$2M in additional annual revenue and retaining ~$800K in at-risk annual business.
Reduced clinical-study workflow cycle time from three weeks to 30 minutes. Urine-sediment models exceeded 96% sensitivity and specificity, supporting FDA clearance, IVDR certification, and international market access.
Scope. Contractual work on Anthropic’s external team through Turing, including LLM evaluation and tool-use tasks.
Delivery. Evaluation of model behavior on tool-use tasks for a frontier AI program.
B.Tech, Electronics & Communication Engineering
I grew up in Warangal and now live in Hyderabad. I studied electronics at IIT (ISM) Dhanbad, taught myself statistics and data science, and started working in the field in 2019.
My career has connected technical work to commercial outcomes: diagnostic devices reaching new markets, enterprise automation generating revenue, and vision systems becoming products. That experience now shapes how I build ClutchLabs.
I’m happy to discuss applied AI, research collaborations, and challenging computer vision problems.
ashwath.basani@gmail.com