Work

Anonymized engagement patterns from real programs I led or architected. Figures are order-of-magnitude.

BFSI

Engineering

~75 % fewer
False positives: >3× more
Novel fraud caught: ~70 % less
Investigation time

LLM · Anomaly Detection · Streaming

Adaptive Fraud Detection Network

Challenge

The legacy rule-based fraud engine flagged overwhelmingly legitimate transactions, so investigators spent their days triaging noise while novel fraud patterns slipped through. Losses were rising despite a growing team.

Approach

Replaced static rules with an adaptive detection network: streaming feature pipeline, anomaly models with LLM-assisted case summarization, and a feedback loop from investigator decisions back into the models.

Impact

Investigators now start from ranked, pre-summarized cases instead of raw alerts. False positives fell by roughly three quarters, materially more novel fraud is caught, and triage time per case collapsed.

Architecture

~80 % faster
Onboarding time: Zero
Audit findings: ~5× more
Analyst throughput

KYC Automation · Regulatory AI · Entity Resolution

Automated KYC & Enhanced Due Diligence

Challenge

Onboarding a corporate client meant analysts manually reviewing hundreds of pages of identity documents, ownership structures, and sanctions lists. The average cycle ran to three weeks, and inconsistent risk assessments across regional teams forced frequent rework. The result was regulatory exposure alongside client frustration.

Approach

Built a multi-agent KYC pipeline that extracts entity data from submitted documents, resolves ultimate beneficial ownership chains, and cross-references global sanctions and PEP lists in real time. A human-in-the-loop stage reviews edge cases while low-risk profiles are auto-approved.

Impact

Onboarding that once took weeks now completes in days, analyst throughput rose several-fold, and the subsequent audit cycle returned no regulatory findings.

Automotive & Manufacturing

Architecture

>£1 m
Value created: <1 hr
Repair SLA: ~65 % less
Avg handle time

Multi-Agent System · Knowledge Retrieval · Service Ops

Technician Support Agentic Swarm

Challenge

Growing complexity in vehicle electronics pushed diagnostic times up, with technicians spending hours searching fragmented PDF manuals. The bottleneck caused missed SLA targets and lower first-time fix rates, which hit dealership profitability.

Approach

Built a multi-agent system, with intake, diagnostic, and repair agents, that automates root-cause analysis. The agents retrieve knowledge from more than 22,000 manuals and guide technicians through repairs step by step.

Impact

Average handle time fell by roughly two-thirds, repairs resolve within the SLA window, and first-time fix rates improved, together generating over a million pounds in value.

Logistics & Supply Chain

Architecture

~95 %
On-time delivery: ~30 % fewer
Empty miles: >€1.5 m
Annual savings

Fleet Optimization · Predictive Logistics · Real-Time Dispatch

Predictive Fleet Dispatch & Delivery

Challenge

Route planning depended on dispatcher experience and static schedules, leaving fleet capacity underused. Driver availability, traffic, and delivery time-windows sat in disconnected systems, which caused frequent missed SLAs and excessive empty-mile runs that inflated fuel costs.

Approach

Built a predictive dispatch platform that combines real-time telematics, weather, and traffic data with an optimization core. Loads are assigned by predicted ETA, hours-of-service compliance, and vehicle capacity, and routes rebalance as conditions change.

Impact

On-time delivery climbed into the mid-nineties, empty miles dropped by roughly 30 percent, and the network now saves over one and a half million euros a year.

EPCM & Construction

Engineering

~60 % faster
Review cycles: ~80 %
Checks automated: Zero
Compliance misses

Generative Design · BIM/IFC · Compliance

Generative BIM Compliance

Challenge

Cross-referencing building information models against evolving local safety codes was manual and error-prone. Engineers spent weeks on visual inspection, which created a bottleneck in the design phase and raised the risk of costly retrofits during construction.

Approach

Built a geometry-aware engine that parses Industry Foundation Classes (IFC) files and validates building models against local safety codes automatically. Checks that took weeks now run in minutes.

Impact

Review cycles run around 60 % faster, the majority of compliance checks are automated, and no compliance misses have surfaced.

Strategic AI & Enterprise

Strategy

>40
Execs upskilled: >20
Use cases launched: ~90
Program NPS

Executive Strategy · Change Management · AI Adoption

VP-Level AI Academy

Challenge

The leadership team had access to capable AI tools but no framework for integrating generative AI into the product roadmap. The gap stalled innovation, as decision-makers struggled to identify use cases beyond simple efficiency gains.

Approach

Designed a “GenAI in Product” executive program that coaches CXOs on roadmap development through a series of workshops, connecting technical capability to business strategy.

Impact

More than forty leaders completed the program, over twenty high-value use cases moved into launch, and it became a blueprint for enterprise-wide adoption.

TOOLING · SELECTION