Work that sits at the boundary

Natural-language SQL agent for non-technical users in regulated healthcare environments.

Healthcare and other regulated organizations sit on data that only a handful of analysts can actually query. Athena closes that gap: it lets non-technical staff ask questions in plain language and get back real answers from the underlying database, without writing SQL or waiting in a request queue.

Multi-cloud infrastructure discovery and Terraform generation, with a FinOps cost layer.

Every cloud environment eventually splinters into three disconnected sources of truth: the wiki that's out of date, the console that shows what exists, and the bill that shows what it costs. Terracarta discovers real infrastructure across AWS, Azure, and GCP and generates Terraform directly from it, then layers in a FinOps module — cost KPIs, waste detection, and CUR 2.0 ingestion — so cost and infrastructure live in one artifact instead of three.

A first-of-kind Ignition SCADA module bringing vector similarity search natively into the platform.

SCADA platforms and modern AI infrastructure don't normally meet. This module changes that: it exposes vector embeddings and pgvector similarity search as a native `system.pgvector` scripting namespace inside Ignition itself, rather than requiring an external service call. It ships with a showcase landing page and an automated GitHub Actions distribution workflow.

Forecasting functions for Ignition 8.3, validated with a purpose-built Jython test harness.

Adds forecasting directly into Ignition's scripting surface — `forecastLinear`, `forecastHolt`, and `forecast` — backed by a Jython test harness built specifically to validate them, with per-step error tracing and a documented set of forecasting limitations rather than a black box.

Two architecture case studies: predictive maintenance at the edge, and institutional knowledge with access control.

DIANA (Distributed Intelligence Adaptive Network Architecture) is a three-tier Azure IoT platform — edge, gateway, cloud — for predictive machine maintenance, with federated learning, OTA updates, and fault explanation built in. Tribal takes a different problem: cross-department institutional knowledge, gated behind four sensitivity tiers with attribute-based access control, so an AI system can answer questions without leaking what it shouldn't.