Things I build to think
These are my own side projects — self-initiated and built end-to-end to test ideas and prove out approaches. They're what I can show openly. My professional and client work is confidential; I'm glad to walk through that in person or under NDA.
APPLIED-AI PROJECTS
SELF-INITIATED · BUILT END-TO-ENDEnterprise-Style AI Assistant
A self-built generative-AI assistant on Azure — semantic and vector search over knowledge sources, retrieval and response workflows, and content-management capabilities. I built it the way an enterprise system needs to be built: the hard part wasn't the chat, it was the production envelope around it — identity, networking, logging, data access, reliability, and supportability.
RFP Response Intelligence
A system for ingesting complex request-for-proposal documents, extracting the requirements and questions, searching organizational reference material, generating draft answers with source traceability, and assembling a response in the client's requested structure. More than a writing tool — a way to capture and reuse organizational knowledge while keeping humans in ownership of the final response.
PDF deconstruction, requirement extraction, and parallel question processing.
Search across prior proposals and project references; draft answers with cited sources.
Freshness, confidence, and reuse controls; user, team, and org preferences; approval workflows.
Clinical Notes & Reporting
A tablet-first app that helps clinicians locate patients, synthesize clinical notes, and generate different report types — with inline editing, highlight-and-comment, iterative rewriting, quality evaluation, and human review at the centre. Built on simulated clinical data, emphasizing usability and transparency over raw generation.
Manufacturing Operations Architecture
A reference-architecture exploration for plant operations — weighing plant independence against enterprise standardization, and local versus centralized decision rights across SaaS/cloud/on-prem systems. Maps data platforms and Databricks analytics onto AI/optimization use cases: production planning, demand forecasting, quote and specification review, and service-call insights.
Application Modernization Study
A comparison of modernization approaches for a mixed on-prem/cloud estate — hybrid API integration, an enterprise replication/data layer, SaaS moves, and introducing AI without prematurely replacing core systems. Weighed against business impact, effort, quick wins, and avoiding work that would later be discarded.
Legislative Knowledge Graph
Built a structured graph representation of Canadian tax legislation in Neo4j, parsing large legislative XML sources into connected entities — text, paragraphs, subsections, and their relationships. A demonstration of turning complex source structures into systems that support navigation, retrieval, reasoning, and future AI applications.
AI-Driven Sales & Workflow Automation
Designed AI-assisted sales and automation platforms — Python services, PostgreSQL, workflow automation, data ingestion and transformation from external sales-data sources, voice agents, and prompt architecture. Paired with the unglamorous-but-essential parts that make automation responsible: compliance processes, do-not-contact controls, and operational monitoring.
INDEPENDENT PRODUCTS
BUILT END-TO-END7 Degrees
Turns every hallway chat, business card, and voice memo into shared, AI-organized relationship memory — so your whole BD team knows who knows whom, and why it matters. Capture an interaction in seconds; an AI agent builds Wikipedia-style pages for every person and organization, puts you on a searchable relationship graph, and surfaces warm intros before connections go cold.
Publish any event to a login-free page with a QR code. Walk-up attendees scan, see a live mesh of who's in the room, and add themselves — every self-registration flows straight into the team's graph, attributed to the host.
Legacy
The workout tracker your members actually keep using. Legacy treats the phone as the source of truth and the network as a background helper — every set logs instantly with no spinner, an in-progress workout survives the app closing or the Wi-Fi dropping, and it reconciles silently across every device the moment it's back online.
Weight × reps with previous-set hints, supersets & circuits, rest timer, live PBs, estimated-1RM / volume charts.
Duration, distance, speed, incline, resistance, calories, heart rate — plus cardio PBs.
Full PWA to the home screen, usable offline, start with no account, magic-link or password, multi-user.
Frontier Exchange
A browser-based, real-time, multiplayer economic-strategy game where your body is the cursor. Server-authoritative, up to 12 players per 15-minute round, no install or accounts. A study in systems design: marginal pricing plus fixed-base scoring makes market manipulation self-harming, and because the client sends intents only, anti-cheat is a property of the architecture rather than a bolt-on.
THE FOUNDATION
WHY THE AI WORK HOLDS UPThe AI work sits on a much broader technical career. Earlier, I worked in data leadership and was part of industry discussions around data virtualization and federation — and my experience runs from embedded hardware up through distributed systems and enterprise consulting. That breadth is why I can reason across the whole system rather than treating AI, data, infrastructure, software, and users as separate concerns.
The rest is under NDA
The projects above are what I can share openly. My day-to-day client engagements — enterprise AI, data, and architecture delivery — are confidential. I'm happy to walk through the details, the decisions, and the outcomes in person or under NDA.
Want detail on any of these?
I'm happy to walk through the architecture, the decisions, and what I'd do differently.