Why I built Klarix
The path from agricultural computer vision to competitive intelligence, and why it's the same problem.
Shipping production AI at John Deere
I work as an AI engineer on FurrowVision, one of the largest precision-ag programs in the world. Real-time computer vision on A10 GPU clusters, 38M+ sensor images, F1 from 0.44 to 0.92 across eight months of dataset curation and model compression. That's where I learned how to make AI actually run in production, not just in notebooks.
Three years of building startups in my free time
Since 2023 I've spent every night and weekend building. First came Speaksense, a multi-tenant LLM platform with English-to-SQL co-agents, pgvector embeddings, agency-scoped RBAC, and Stripe billing. That taught me what it takes to ship an AI product end-to-end: auth, billing, observability, and the operational glue most founders skip.
Sales teams flying blind
Every B2B team I talked to had the same complaint. Paying $5K+/month for Apollo, ZoomInfo, Clay, Gong, getting raw data they couldn't use. Reps burning 20+ hours a week researching prospects, still walking into deals blindsided by competitors they'd never heard of. Marketing had a battlecard nobody opened. Mandy McEwen calls it Battlecard Theater. The intel existed. It just wasn't in the rep's hand at the moment of truth.
Intelligence as a deliverable, not a dashboard
I built Klarix because the market didn't need another tool, it needed an outcome. Dr. Nici Sweaney's framing for AI fits exactly: it's an operating system underneath, not another app to adopt. Dossiers, battle cards, SWOTs, surface maps, outreach sequences. Ready to use. Delivered in 3 to 7 days. Backed by a human-gated, multi-source pipeline with cost-optimized model routing, failover, and citation gates — the same engineering rigor I apply at the day job, pointed at competitive intelligence.