2026 · Independent Applied Research & Development
From complex problems to working systems
Three projects spanning product development, institutional research infrastructure, and independent AI evaluation.
Each began with a different problem and required a different approach. Together, they demonstrate how I define requirements, direct AI-assisted implementation, coordinate research and testing, evaluate results, and turn findings into decisions.
The work includes functioning software, governed data structures, repeatable operating processes, and controlled experiments. Each case study provides a concise account of what was accomplished, followed by the architecture, evidence, and decisions behind it.
Market Intelligence · Product Development · Research Operations
Market Weather
Environmental intelligence that helps traders understand changing market conditions.
Market Weather was built for traders who need to read changing conditions as a whole rather than as a wall of separate indicators. It translates volatility, credit, rates, liquidity, and related market information into understandable market environments.
I directed the development of a working market-intelligence prototype combining a Streamlit Forecast Center, market-regime classification, participant archetypes, historical replay, and an AI-assisted research system.
The distinctive innovation was connecting the product to a governed learning process. Research observations could become investigations, historical comparisons, and qualified institutional memory—but only through explicit review and evidence requirements.
WHAT THIS WORK DEMONSTRATES
• Product development: Directed a functional market-intelligence interface and the research infrastructure supporting it.
• AI-assisted research operations: Established structured collaboration, mandatory dissent, research assignments, and evidence-based review.
• Operational governance: Directed the rollback of an entire day’s analysis, with individual registry reversals verified and recorded.
• Evidence-driven leadership: Kept adverse findings in the record and required human adjudication, so interpretations were left provisional, corrected, or rejected when the evidence did not support them.
Implementation evidence: 484 commits · 1,204 tracked files · 779 test functions · 556 design documents.
Status: Working research prototype.
Systems Design · Requirements Engineering · Program Delivery
Sentinel
Institutional knowledge that doesn’t depend on somebody remembering what happened.
Organizations accumulate questions, decisions, evidence, exceptions, and unfinished investigations across people, departments, and systems. Too often, the relationships among those records disappear as work progresses.
Sentinel explored a different approach: a persistent institutional research system where evidence, decisions, authority, and unresolved questions are represented as governed, traceable objects.
I directed an AI-assisted software implementation using PostgreSQL, database constraints, structured research objects, and controlled workflows. The objective was to move governance beyond policy documents and into enforceable system behavior.
WHAT THIS WORK DEMONSTRATES
• Requirements and systems design: Translated institutional research and governance requirements into an implementable object model.
• Technical delivery: Directed an implementation with four Alembic migrations defining 19 database tables.
• Quality assurance: Oversaw automated testing, with 1,082 test functions across 81 test files in the verified repository.
• Governance and evaluation: Applied segregation of duties, controlled write authority, and traceable decision processes; implementation reviews produced documented correction work on governance behavior and research workflows, while credential isolation and a bounded real-world pilot remain outstanding.
Status: Experimental software implementation; not deployed to institutional users.
AI Evaluation · Adversarial Testing · Independent Verification
Cairn / Semantic Integrity Audit (SIA)
An independent audit caught what an AI agent failed to disclose.
AI systems can describe their own reasoning and behavior, but those descriptions are not independent evidence of what occurred.
Cairn investigated a specific question: could an AI system’s interpretation of a task—and its account of that interpretation—be examined independently?
The Semantic Integrity Audit (SIA) was the auditing mechanism. The experiment used sealed conditions, independent reviewers, structured adjudication, and preserved evidence to compare the acting agent’s disclosures against independently examined behavior.
WHAT THIS WORK DEMONSTRATES
• Experimental design: Directed a controlled evaluation with defined conditions, independent auditors, and evidence-handling requirements.
• AI evaluation: Tested 252 audit units through 504 independent auditor exposures, with 892 adjudicated pairings.
• Independent verification: The acting agent disclosed its intention to reinterpret in 195 of 196 relevant instances; the audit detected the single undisclosed exception.
• Research governance: Maintained the distinction between a voided initial experiment and the valid Experiment 1.1 result, preserving the evidentiary limits of the finding.
Status: Bounded experimental research with a documented mechanism failure; the audit mechanism is not a validated commercial product.
The common thread
These projects are different in purpose, but they share an operating discipline: establish the problem, define what success requires, build and test the mechanism, and allow the evidence to change the decision.
That approach draws on my background in institutional technology, enterprise client management, operational leadership, and cross-functional delivery.