One résumé. Two lenses.

The business needed AI that teams could actually use without creating new risks. I architected the platform layer that made that possible: a governed skills engine (800+ skills), a model gateway abstracting across Claude, GPT, and Gemini, and a guardrail system that catches unsafe outputs before they reach users. Business teams build their own AI capabilities on top — the platform enforces the rules so they don't have to think about them.

The hardest part wasn't the tech — it was designing a system where the safe path is also the easy path. That's the work I'm most proud of here.

Field expenditure at an oil major is opaque by default — invoices, vendor contracts, and spend data living in separate systems with no single view. I designed and built a full-stack monitoring platform that consolidated that data, surfaced anomalies, and gave decision-makers a live view of where money was going. The business case paid for itself before the year was out.

HSE teams were spending most of their time managing data instead of managing safety. I redesigned the underlying data model for compliance tracking — connecting well, lease, and facility data into a coherent regulatory graph — and built the pipelines that kept it current. The team shifted from reactive data cleanup to proactive risk management.

The problem we set out to solve. Every oil & gas operator in Texas lives under the Railroad Commission (RRC), and the cost of regulatory compliance is quietly enormous. Routine actions — drilling a well too close to a lease line (Rule 37 spacing exceptions), flaring or venting gas (Rule 32), monthly production reporting (Form PR) — each require specialized filings prepared by scarce, expensive talent: regulatory analysts, landmen, and attorneys. The work is manual, deadline-driven, and unforgiving. A missed deadline can halt production. A flawed flaring justification can trigger fines and emissions liabilities. As ESG and methane scrutiny intensify, the penalty for getting it wrong is rising while the experts who get it right are retiring.

What AEGIS is. AEGIS is an AI compliance platform that turns this reactive, headcount-bound burden into a proactive, monitored system. It ships four specialized AI agents — for spacing exceptions, flaring exceptions, portfolio-wide compliance monitoring, and flaring/emissions tracking — that continuously watch an operator's wells, surface risks before deadlines hit, and assemble regulator-ready filings.

Why it can be trusted in a regulated business. The reason compliance hasn't been automated isn't capability — it's trust. AEGIS is built for it: every regulatory filing passes through a mandatory human-in-the-loop approval before it goes anywhere, and every action is written to a tamper-proof audit trail. The AI does the heavy lifting; a qualified human signs off; the regulator-grade paper trail is automatic. It augments the expert, it doesn't replace the accountability.

The business value. AEGIS compresses filing cycles from days of expert time to minutes of review, lets a single analyst cover a far larger well portfolio, and shifts compliance from "scramble before the deadline" to "flagged weeks in advance." It reduces the two costs that hurt most: the labor cost of producing filings and the financial/reputational cost of getting them wrong.

The market. This is a wedge, not a ceiling. The architecture is a general stateful runtime for governed enterprise AI agents — the same trust-and-audit foundation extends to other Texas rules, other states, and ultimately any high-stakes, regulated workflow where AI must be both capable and accountable.

The problem I saw at a client. Working with an enterprise rolling software out to more than 5,000 users, I watched every new application — and especially every new AI agent — trigger the same manual cycle: trainers, scheduled sessions, static decks, reactive tickets. Conversational AI made it worse: there's no fixed UI to memorize, and users need coaching in the moment, not documentation after the fact. Adoption stalled and the software investment sat underused.

What EVIA is. An adoption platform that treats video as a machine-understood artifact, not passive content. A user records a problem or a task; EVIA analyzes it semantically — not just transcription — to infer the application, the UI state, the actions, and the intent; then it returns video-first guidance assembled from a curated knowledge base. When the issue is a confirmed defect rather than a knowledge gap, it's bundled with full context and routed straight into the enterprise's existing ticketing system.

What I own. I authored the full vision end to end — architecture, user personas, integration surface, governance and privacy model, and strategic positioning — ran a competitive scan that confirmed no one solves the integrated problem, and I'm personally building the MVP. Retrieval starts on full-text search, with a hybrid architecture designed to scale to embeddings and LLM re-ranking as the corpus grows.

Why it matters now. AI-agent rollout is the beachhead: it's where traditional enablement breaks down hardest and where adoption — not licenses — decides whether the investment pays off. EVIA is positioned to compress enablement cost and time-to-productivity for exactly the organizations betting on agents at scale.

This is pure research curiosity, driven by the observation that guardrails are becoming mission-critical for enterprise AI but most solutions are either black-box APIs (no visibility, no control) or require expensive custom annotation. The hypothesis: Anthropic's constitutional-classifier approach — synthetic data generation at scale, trained lightweight binary classifiers, deployed locally — could be adapted as an enterprise pattern.

The experiment: define a constitution (rules for what is/isn't safe in a specific enterprise context), generate synthetic positive/negative examples, train small classifiers on the synthetics, evaluate against real refusals and false positives. If successful, it becomes a defensible alternative to vendor lock-in for safety infrastructure. Currently in the exploration phase.