I lead product, design and delivery for complex, data-heavy platforms.

The problem

Analysts managing multi-asset portfolios worked across four disconnected tools: a risk system, a transaction feed, a spreadsheet tracker and a reporting module. They spent 2–3 hours every morning assembling data before any analysis could begin.

What I did

• Owned requirements, prioritisation and the roadmap, presenting direction to leadership across 5+ planning cycles.
• Ran sprint and release planning and estimation sessions across design, frontend and backend workstreams.
• Specified 30+ user flows for portfolio analytics, risk monitoring, transaction visibility and decision dashboards.
• Defined every production data scenario with backend engineers before build.
• Led design from a blank canvas to a production design system that let engineering build modules in parallel, and mentored 2 designers through weekly reviews.
• Turned regulatory and compliance requirements into 40+ testable acceptance criteria across 3 modules.

Results

• 20% faster design to development
• 1–2 weeks of rework saved per release
• ~15% fewer revision cycles
• ~1 week faster decisions per planning cycle
• 0 design-related compliance flags at QA

The problem

Manufacturing plants ran production, workforce, inventory and reporting on separate systems. Supervisors lacked a real-time view of the floor, so decisions came late and problems surfaced after output was already lost.

What I did

• Defined a role-based platform covering a supervisor’s command centre, a live floor monitor, an equipment registry, incident and work-order management, and operations analytics.
• Designed the real-time floor monitor for 12 production lines: cycle time, yield, output against target, live line status and a 24-hour production timeline.
• Structured incidents as a tracked lifecycle (detected, investigating, assigned, resolving, verified, closed), linking sensor alerts, root cause, affected lines and work orders.
• Shaped analytics that trace output from theoretical capacity to actual (a loss waterfall), split downtime by cause and rank equipment by financial loss.

What each role gets

• Supervisors see the anomalies that need action first
• Maintenance gets prioritised work orders tied to alerts
• Managers see OEE, downtime and the cost of each loss

Select a screen to enlarge it. Screens show sample data.

The problem

Revenue and industrial development officials across a state’s districts needed structured access to land records and programme data that sat in fragmented systems.

What I did

• Defined product requirements, information architecture and user journeys for a multi-role platform.
• Tested prototypes iteratively with department officials and used the findings to shape navigation and workflows before rollout.
• Ran the programme alongside other concurrent enterprise engagements, handling planning, estimation and stakeholder coordination.

Results

• 10,000+ public-sector users given structured data access
• Workflows validated with end users before rollout
• Delivered on time

The problem

The client wanted a rules-based system for CME equity-index futures, and needed evidence that it would hold up under strict risk limits before any capital went live.

What I did

• Defined a layered architecture: market-data ingestion, analysis engine, rule-based decision engine, execution and a monitoring dashboard.
• Planned a phased rollout from backtest to paper trading to controlled live release.
• Built a validation pipeline on 16 years of 1-minute CME market data, with hard go/no-go gates on profit factor and drawdown.
• Built in risk controls: per-trade, daily and trailing drawdown limits, plus an exclusion calendar of 524 economic-event days (FOMC, CPI, NFP).

Results

• 5.4M+ rows validated, zero data-quality issues
• A failing version was stopped at the gate and redesigned
• Each risk rule’s effect measured before adoption

The business question

A software company’s sales, marketing and support data sat in three separate systems. Leadership couldn’t see how campaigns turned into revenue, or which valuable accounts were at risk.

What I did

• Aggregated each source (499 deals, 2,240 campaign records, 3,000 support tickets) to account level and joined them on account ID into one 65,448-row, 63-column model.
• Created the measures leadership asks about: Days to Close, Campaign ROI and Resolution Time.
• Built three linked views (Marketing Funnel, Sales Pipeline, Customer Health) with filters and cross-sheet navigation, and wrote the finding and recommended action into each.
• Set up plain-English questions on the data with a Quick Topic, and checked its answers instead of trusting them.

What it told leadership

• Every channel had negative average ROI, so spend needed reallocating
• One account combined $40,722 in deals with the most support tickets (334)

The business question

A B2B AI software company needed to understand the 2026 AI tools market (competition, pricing and adoption) to decide how to position its product.

What I did

• Framed the research before querying: objective, scope, exclusions, competitor set, timeframe and constraints.
• Configured a no-code AI agent on the internal dataset and combined it with external Quick Research.
• Required every finding to say whether it came from the internal data, the external research or both.
• Rated each insight for source quality, consistency, timeliness and confidence, and wrote down what the evidence could not support.

What it told leadership

• Five evidence-backed insights in a leadership brief
• Pricing is moving from per-seat to consumption-based
• AI tools’ capability is ahead of enterprise readiness

Select a screen to enlarge it. Both projects use sample data.