A Houston engineering practice for Epicor Prophet 21 and Kinetic: implementation, data migration, integration, rescue, and managed support, plus an AI platform purpose-built for distribution and manufacturing ERP.
The next generation of ERP is AI-native. We're building it, layer by layer, on top of the systems distributors already run.
Fine-tuned models trained on your ERP data: schema-aware, transaction-literate, and deployed local-first so your data never has to leave.
Autonomous AI workers that handle the busywork inside Prophet 21 and Kinetic, verified, gated, and auditable at every step.
An AI-powered phone system with live transcription and in-call intelligence, wired straight into the ERP your team already uses.
Every one starts with a conversation with an engineer, not a sales sequence.
Everything we ship is built by Anvil, our in-house autonomous engineering system. A long-running coding agent with oracle-verified writes, paired-reviewer gates, semantic memory, and a fleet runtime that scales from a single laptop to an 8×B200 swarm. Nothing counts as done until a mechanical verifier agrees.
A specialized model and the open benchmark it climbs. Trained without a single human label: every preference pair was graded by executing candidate queries against a live SQL Server.
Text-to-T-SQL for distribution ERP, tuned with execution-grounded DPO. Beats its base model on every split, with the largest gain on the private holdout. Ships with schema-constrained decoding so it cannot invent a table name.
An execution-graded text-to-T-SQL benchmark on an original 60-table wholesale-distribution schema: 2.56M seeded rows, 6,200+ verified questions across five difficulty tiers, and the full grading harness so results reproduce from a clean checkout.