The Asymmetry Problem
Why Human–AI Trust is Fundamentally Broken
Builder. Problem Solver. Pragmatist.
I'm Vishal. I've helped build and ship 25+ tech products in the last 10 years. From complex AI systems detecting weapons at stadiums to private apps that find and encrypt your nudes before someone else does. I architect, code, and lead teams that turn complex problems into tenable solutions. When I'm not building, I'm shooting film, collecting vinyl, or writing about philosophy, tech, and more.
I cofounded Labyrinth in 2017 and have been building and researching in machine learning and artificial intelligence since. Along the way, I also cofounded Verizy, built Cacti among many other products, and fostered a team that thrived through the good and the hard times.
Latest
What I've Built
Products & Companies
Labyrinth BootstrappedAIML
Co-Founded and led this AI/ML consulting company building production ML and AI systems for small scale and enterprise clients. Auto-healing, automatically scaling, fault-tolerent, custom-trained, ML pipelines, and models deployed at scale-up production systems with authenticated first-class API and queue based access.
30+ clients | 20+ team members | Revenue in the 7 figures | 8+ years in operation
Verizy SAASMLIdentity Verification
Co-founded & architected the AI-powered identity verification platform for KYC compliance. Tech that verified millions of users across India.
10 million+ verifications | 99.2% accuracy | Used by various large organizations
Akuri SAASAILLM
Context-aware AI that lives in your text boxes. No more copy-pasting from ChatGPT. Give AI features to all your customers in under a day.
Cacti AppPrivacyEncryptionML
On-device ML finds and vaults your sensitive photos. 256 bit AES encryption. Zero cloud. Your secrets stay yours.
10s of thousands of users | 4+ years in operation
Selected Work
Orchestral Knowledge GraphExplainable AIAI for Data Center
Explainable AI system for managing data center and on-prem infrastructure. It detects faults, triggers automated remediation via StackStorm, and enables preemptive reprovisioning to reduce downtime. A graph database maps & tracks millions of nodes, with ML models predicting blast radius, mitigation strategy, & timelines.
Xonar Sensor FusionData WarehouseOutlier Detection
High-throughput weapon detection system using AI to analyze radar, LiDAR, and camera data. It identifies object shape and density, filtering out common non-threats while detecting weapons. Work included anomaly detection, outlier handling, retraining workflows, and centralized data warehouse management.
Circa Graph DBAI for Supply ChainClimate Tech
Industrial production systems and supply chains modeled as graphs of materials, processes, and metrics. Neo4j. LCA and proprietary analyses optimize the chain for energy, cost, time, and footprint. ML models trained on this graph data provide optimization suggestions and insights.
Tagwalk PythonElasticsearchComputer VisionReact
Hierarchical multi-label image classification system for tagging fashion industry photos. Trained on millions of images across 8 models working in unison to identify material, pattern, color, style, vintage, and more—covering over 2,000 labels. Includes weekly automated retraining with a champion-challenger, human-in-the-loop workflow. Powers visual search and similar outfit recommendations.
Reduced manual tagging time by 93% | Tagging error reduced from 26% to 2%
Decision Review System 🤐 In StealthSports-techAIComputer VisionML
AI-powered DRS for sports. The tech that helps umpires make the right call. High-throughput, real-time, high frame rate, custiom trained AI models, multi-camera triangulation.
ICC Immersive Vision ProDigital TwinLive VR StreamingComputer VisionARVR
Live player position, ball trajectory and physics captured from live Cricket World Cup matches using specialized cameras. Digital twin recreation of the matches in Apple Vision Pro app with digitally modelled stadium, players, and ball. Experince the match as the batter with real ball speed, physics, and player positions recreated.
Built in association with International Cricket Council (ICC), Apple, AWS, and Quidich.
Discrete Temporal ML ModelSalesperson App with Redacted DataCentral Budgeting for Gifts
Client relationship tool for high-income sales, surfacing personal details (birthdays, interests, etc.) and automating gestures like buying sports tickets through learning from past results that lead to increase in conversions. Used by luxury dealerships to stay top-of-mind of potential and past customers.
Was able to automatically keep client relationships alive with personalized gifts and gestures for a few 100 high networth clients in NYC.
Pltfrm WebSocketsReactSwiftOffline-SyncTime-Series Diff Data
Real-time, Trello-like project management tool for web and mobile. Led tech and built WebSocket architecture to scale with auto ramp-up/down. Used a time-series diff DB for offline & relatime sync.
AGNT Weekly RetrainingMatching AlgorithmKnowledge Graph
Airbnb for DJs and event talent” — built web and mobile platforms with personalized feed recommendations and a matching algorithm connecting bookers with talent based on behavior and preferences.
Born Knowledge GraphPersonalized FeedFull Stack
Multi-tenant, always-on online tradeshow and retail marketplace. Used behavior learning and knowledge graph-based recommendations to personalize feeds and connect buyers with millions of products.
BeLoyal defunct
Full-suite loyalty program with location-based nudges and smart offers triggered by user behavior and events. Led architecture and tech, and designed the nudging system.
Hyperface ClusteringHigh-dimensional DataCredit CardFintech
Dynamic credit card user clustering system for banks and issuers using 100s of data dimensions. Automatically redraws cohorts based on spending patterns to target offers and analyze profitability.
Hobby & Personal
Bonsai Coming SoonApps
App to downsize your large videos to save space on your phone and pay less for iCloud.
Coming soon - Join the beta now!
ControlHub & NextHub Apps
Native dashboards for ControlD and NextDNS. Because web interfaces on phone browsers aren't always convenient.
ControlHub ⭐️⭐️⭐️⭐️⭐️ 4.9/5 | NextHub ⭐️⭐️⭐️⭐️⭐️ 4.9/5
How I Work
Built teams that ship. Established a culture where: Arguments are about problems, not egos; ML projects get discovery trees, not fake deadlines; Engineers talk to clients, not salespeople; Knowledge sharing isn't optional. I believe in building what works, admitting what doesn't, and iterating fast.
Recognition & Media
Featured in YourStory for building scalable AI solutions using AWS infrastructure. Discussed machine learning architecture in an AWS customer spotlight interview. Recognized at WWDC 2019 for innovation in mobile development.
Also
I contribute to entropy. I'm an egalitarian. I make machines learn. I blog at Musings. I founded Labyrinth. Watch my AWS interview or catch me at WWDC. Film photography and vinyl keep me sane.