How AI was applied
Tile size and shade show how many contributions use each AI technique. Select a family, a tile or a cell in the matrix to read the contributions behind it.
Across the career, oldest to newest
| AI technique | Early career 1994–2006 | UN CGIAR 2002–2011 | UN Bonn 2012–2013 | Nespresso 2013–2018 | Fair Trade 2018 | DSM-Firmenich 2019–2024 | Solara28 2025–2026 | EPAM 2024–2025 | Gavi 2026 |
|---|---|---|---|---|---|---|---|---|---|
| AI strategy & governance | |||||||||
| AI strategy & AI Factory | |||||||||
| AI governance & Responsible AI | |||||||||
| AI risk, security & EU AI Act | |||||||||
| AI adoption & literacy | |||||||||
| Generative AI & LLMs | |||||||||
| LLMs — GPT, Llama, Claude, Gemini | |||||||||
| RAG & vector search | |||||||||
| Copilots & conversational AI | |||||||||
| Prompt engineering & governance | |||||||||
| Document & catalogue extraction | |||||||||
| Generative media — images & video | |||||||||
| Machine learning & deep learning | |||||||||
| Deep learning — LSTM, CNN, attention | |||||||||
| Digital twins | |||||||||
| Predictive analytics & forecasting | |||||||||
| NLP & text analytics | |||||||||
| Computer vision & recognition | |||||||||
| Anomaly detection & optimisation | |||||||||
| Reinforcement learning | |||||||||
| MLOps & LLMOps | |||||||||
| MLOps pipelines & model lifecycle | |||||||||
| LLMOps & AI evaluation | |||||||||
| AI observability & FinOps | |||||||||
| Agentic AI | |||||||||
| Multi-agent systems | |||||||||
| Agentic software delivery | |||||||||
| MCP & agent tooling | |||||||||
| Agents on Microsoft 365 & Azure | |||||||||
All AI contributions
81 contributions
2024–202524 contributions
EPAM Systems
- Led the technical response to a Tier-1 Swiss bank’s enterprise AI RFI — structured on business value; technology, security and compliance; and people, operating model and governance — later reused as the technical core of a second pursuit with the same bank and the engagement behind EPAM’s preferred-vendor status.
- Coordinated the proposal for a bank-wide Generative AI blueprint mapping 35 prioritised use cases across all seven divisions of the bank over six enabling layers: AI strategy, legal and compliance, data and security, technical architecture, responsible AI and workforce enablement.
- Contributed the AI and data proposition to a growth proposal for the bank’s China digital wealth platform (200,000+ users in two years, in a market where over 80% of wealth management is domestically held): a gamified paper-trading experience with a graduated three-tier conversion model, disclosure design and an explicit Chinese gaming-regulation risk assessment.
- Global agribusiness and food company, Switzerland: defined the GenAI strategy for shipping and transport optimisation, structuring 18+ use cases (laytime calculations, contract analysis, chartering risk mitigation) with OCR, NLP, LLMs and knowledge graphs.
- Designed the technical questionnaire responses on model orchestration, multi-LLM integration, RAG, embedding strategies and secure deployment, articulating EPAM’s LLMOps approach, AI Factory and compliance measures, and coordinated contributions from global EPAM teams.
- Adapted EPAM’s Generative AI Factory target architecture for the bank — three planes (Models / Data / Services) over Data Factory, AI Backoffice and AI Serving Platform — covering feature and prompt engineering, continuous training and evaluation, data- and model-drift monitoring, hallucination management, guardrails, vector stores, FinOps and a gateway to third-party models, closing on a dedicated data-privacy layer.
- Major US county government: as principal AI/GenAI expert, spearheaded the response to an RFP for enterprise-wide AI and automation services — designed the global delivery model, defined the AI Factory approach and translated requirements into modular solutions for GenAI, automated document processing and operational decision support.
- Global pharmaceutical company: principal expert (GenAI, architecture, software engineering) supporting the AI Factory framework and global AI strategy; defined a reusable AI component framework and LLM/multi-agent orchestration for biomedical R&D and regulatory processes, and a global Community of Practice for AI knowledge sharing.
- Global energy leader, Saudi Arabia: designed scalable AWS AI cloud platforms and end-to-end architectures across upstream, downstream, R&D, engineering, finance, HR, safety and supply chain, with LLM orchestration, multi-agent systems, RAG and LSTM models for R&D.
- Leading telecom operator, Africa: co-led the RFP response on enterprise AI/GenAI strategy, developing a target-state federated AI architecture, AI Factory delivery capabilities and a GenAI integration roadmap.
- Global chemical manufacturer: lead architect for GenAI in R&D, engineering and production — semantic ingestion pipelines, RAG over structured and unstructured data, and a phased path from PoCs to multi-agent assistants on a modular, scalable architecture.
- Global industrial materials firm, UK: defined the architecture for freight anomaly detection — hybrid statistical, ML and GenAI models with RAG for contextual validation on AWS and GCP, feature engineering, prompt design for anomaly justification and EPAM AI accelerators.
- Leading logistics and postal group, UK: designed a strategic MLOps assessment offering for enterprise-wide AI/ML adoption, with an evaluation methodology across data ops, model management and platform readiness.
- Global coffee company, Netherlands: led the analytical assessment of predictive analytics for global shipment delays — historical shipping and customs data, key delay drivers, model performance — and delivered the resulting AI strategy and offering.
- Led an AI regulatory readiness assessment aligned with the EU AI Act, identifying system classifications, governance gaps, documentation requirements, human-oversight controls, transparency obligations and post-deployment monitoring needs.
- Established model-risk controls for predictive and generative AI, including independent validation, bias testing, explainability review, performance thresholds, approval gates, audit trails and periodic revalidation.
- Defined security controls for LLM and AI solutions: identity and access management, secrets management, encryption, network segmentation, prompt-injection defences, content filtering, audit logging, vulnerability management and secure SDLC.
- Established a production MLOps operating model: experiment tracking, feature management, model registry, reproducible pipelines, CI/CD, automated testing, deployment controls, rollback, monitoring and scheduled retraining.
- Defined an LLMOps evaluation framework for RAG and agentic systems measuring retrieval relevance, groundedness, factuality, hallucination rate, task completion, latency, cost per interaction, safety and user satisfaction.
- Set service-level objectives for production AI services: availability, latency, extraction accuracy, retrieval quality, evaluation scores, deficiency handling, incident response and recovery time.
- Implemented end-to-end observability across data pipelines, retrieval layers, LLM calls, agent workflows, model outputs, user feedback, infrastructure performance and operational cost.
- Introduced AI FinOps to monitor and optimise cloud spend, GPU consumption, token usage, model routing, embedding costs, vector-store capacity and cost-to-serve by product or client.
- Defined AI incident-management procedures for model degradation, data-quality failures, hallucination events, security incidents, service interruptions and regulatory escalation, with post-incident review and corrective-action tracking.
- Implemented formal model-lifecycle governance: versioning, approval, retirement, evaluation refresh, data-drift detection and policy updates for predictive models and LLM applications.
202612 contributions
Gavi, the Vaccine Alliance
- Established and operates the Innovation Lab’s experimental AI tooling estate on Microsoft Frontiers models in Azure — including Cowork, Copilot Claude, MS Scout and OpenClaw for agentic-AI experimentation and architecture evaluation, alongside Cursor, Lovable, Bolt, v0, Microsoft 365 Copilot, Copilot Studio and Azure AI Foundry — under defined tool-selection rules, cost guardrails and an approved/excluded tooling list.
- Leads AI enablement through a three-tier training architecture, a curated prompt library, an intranet AI Self-Service Hub and an organisation-wide AI Community of Practice with departmental champions — targeting 80%+ awareness-training completion and 70%+ monthly active Copilot users in year one.
- Covered visitor and case lifecycles, category taxonomies, referral routing and service metrics under hard confidentiality controls — an immutable audit trail, a strict “Chinese wall” between functions, visitor numbers with no names stored and a <5 anonymisation threshold on all reporting — modelled in C4 and validated by an independent penetration test before go-live.
- Designed the Azure architecture extending Gavi’s Programmatic Dashboard from reactive reporting to predictive early warning, forecasting country × vaccine stock-buffer breaches 6–12 months ahead — integrating SAP, UNICEF SD/Thrive360 and VaxNeed into Azure SQL and a row-level-secured Power BI model, with forecasting on Azure Machine Learning and grounded narrative generation on Azure OpenAI behind an API Management LLM gateway.
- Coordinated architecture-board acceptance of the Governance Bot: a permission-aware knowledge assistant for audit-committee, board and governance-decision queries in Teams and SharePoint on Azure AI Foundry, with an eight-metric production-approval framework — answer accuracy ≥90%, citation coverage 100%, fallback precision ≥95%, access-control integrity 100% — so an AI service is not approved on technical completion alone.
- Skills and technologies: Azure solution and security architecture (App Service, Static Web Apps, Functions, SQL, Blob, Key Vault, API Management, Data Factory, Machine Learning, AI Foundry, AI Search, Monitor) · Snowflake · Azure OpenAI, RAG, multi-agent and agentic AI · MLOps/LLMOps and AI evaluation · AI governance, Responsible AI and EU-aligned AI risk management · Entra ID, RLS, TDE/CMK, PIM, managed identities, TLS · C4 and TOGAF · Power BI, Power Platform, Copilot Studio, Purview, Sentinel · confidential-data governance · innovation portfolio governance, vendor selection and procurement · Design Thinking and rapid prototyping facilitation.
- Principal AI Architect and Innovation Lab Coordinator consultant at Gavi, the Vaccine Alliance, leading the AI architecture practice behind global vaccination programmes; owns the end-to-end innovation lifecycle — idea, experimentation, outcomes, stabilisation, operationalisation — as operational deputy to the Innovation Lab Lead, and leads architecture definition and deployment across the AI portfolio.
- Manages the Innovation Lab governance model, structuring innovation into three stepped streams (Ideation → Experimentation → Stabilisation) with five accountable roles, RACI ownership per activity, entry and exit criteria, and a rhythm of weekly experiment reviews, monthly idea reviews and quarterly approval sessions.
- Operationalised AI governance controls covering data leakage, agent proliferation, cost overruns, prompt injection, hallucination and shadow AI, with per-agent monthly credit limits and threshold alerting, DLP connector policies blocking third-party connectors, and monitoring across Power Platform Admin Center, Microsoft Purview and Microsoft Sentinel, reported monthly to the AI Steering Committee.
- Established the reusable reference architecture for grounding Copilot agents in enterprise knowledge — SharePoint → Azure Blob → Snowflake → Azure AI Search vector index → Power Automate → Azure OpenAI → Copilot in Teams/SharePoint — secured with Entra ID RBAC, Key Vault and Purview DLP, with mandatory citation traceability and token-consumption cost controls.
- Develops AI and solution architectures in full collaboration with the information-security and compliance function, as a formal pre-review and clearance gate rather than an after-the-fact sign-off; controls such as Entra ID SSO, least-privilege service access, audit logging of queries, responses and data access, restricted-content protection, Key Vault, PIM, RLS and Purview DLP are specified at design time.
- Defined a multi-tier Governance Copilot ecosystem separating the user-facing Governance Bot, a Governance Knowledge Hub and a Power Automate + Azure OpenAI Query Executor, with Azure Blob ingestion, Azure AI Search vector indexing, Entra ID single sign-on and Azure Monitor / Log Analytics monitoring of latency, token usage and errors.
2025–20268 contributions
Solara28
- Produced promotional videos with AI video generation (Seedance 2.5 in Dreamina).
- Architected and implemented a governed Agentic Factory coordinating eight specialised virtual teams and 32 autonomous AI agents across the end-to-end software delivery lifecycle, generating 6,000,000+ lines of production-ready code, documentation and change artefacts in under one year across a polyglot stack (TypeScript, Supabase/PostgreSQL, FastAPI, Stripe, Vercel).
- Established Solara28 as a large-scale reference implementation of Agentic AI product engineering, demonstrating the viability of autonomous multi-agent systems in enterprise-grade software delivery.
- Designed, built and published the open-source Buonaiuto-Doc4LLM MCP server, which auto-detects a project’s library stack, fetches current official documentation and serves it version-aware to AI coding assistants — the RAG infrastructure that kept the agents’ 6M+ lines of code aligned with current, secure architectural standards and eliminated API hallucinations.
- Established an end-to-end autonomous AI delivery model covering architecture design, code generation, testing, security validation, cloud implementation and documentation, with every artefact traceable to requirements, architecture decisions, source-control changes, code review, automated testing, security scanning and release approval; the Agentic Factory executed work across engineering, QA, DevOps, compliance, localisation, marketing, branding, e-commerce and online payments.
- Governed the automated delivery lifecycle — architecture, software engineering, testing, security, DevOps and documentation — as an agentic AI operating model aligned with the EU AI Act.
- Architected a provider-agnostic LLM service layer (TypeScript/Node.js/Supabase) connecting OpenAI, Anthropic and Gemini APIs, with configurable provider settings, prompt governance, conversation logging and health monitoring, powering conversational and voice-enabled commerce (“Talk to Your Shop”), automated catalogue ingestion from PDFs, photos and Excel, and predictive revenue forecasting.
- Embedded operational AI directly into the commerce execution layer: one-click PDF-to-live-store catalogue imports, natural-language conversational assistants, voice-enabled customer ordering via chatbots, a Merchant AI Assistant and predictive CRM analytics built natively into the platform.
2019–20247 contributions
DSM-Firmenich
- Pioneered an AI hybrid digital twin predicting unintended manufacturing-plant shutdowns that reduced downtime by 48% and earned the Best DSM Corporate Research Project award, using deep learning (LSTM, self-attention) on 768 real-time sensor streams and LLMs (GPT-3.5, Llama 2) in LangChain for maintenance-log interpretation.
- Implemented a production-grade MLOps framework for AI-driven lab automation (CNN/NLP chromatogram recognition), achieving €1.5M annual ROI and €1.2M annual savings through workflow automation.
- Ideated and led research on an AI digital twin modelling the statistical distribution of sun-cream UV filters on skin, aiming to replace in-vivo experiments with digital solutions — a projected value of €30M+ per year.
- Coordinated international Corporate Research Program (CRP) activities in Digital Transformation and the Digital Twin Lab — agile pilots for digital disruption, competence development and continuous improvement.
- Led R&D on a new data-science model automating the identification and quantification of chemical compounds in chromatograms with neural networks, outperforming lab analysts in accuracy and precision, in partnership with Loughborough University (UK).
- Led an information-retrieval project governing millions of lab documents with NLP and transformers (BERT, GPT) for keyword extraction and synonym generation, plus an ontology integrated with public data sources to improve research data interoperability.
- Applied validation-by-design for AI-enabled processes in pharmaceutical and clinical environments: requirements traceability, controlled documentation, test evidence, audit trails, change control and role-based approvals.
2013–20182 contributions
Nespresso (Nestlé)
- Senior Managing Consultant with broad scope across digital and data science in Nespresso’s commercial and e-commerce business, focused on integration, forecasting, communication and coordination across global markets.
- Implemented an A/B/n multivariate testing strategy that significantly improved campaign effectiveness and ROI, integrating sentiment analysis and topic modelling for actionable insights.
20181 contribution
Fair Trade Handels
- Achieved a +561% quarterly revenue increase with a data-analytics marketing campaign using predictive behaviour modelling for the 2018 World Cup.
2012–20131 contribution
United Nations
- Designed and documented a UN institutional data-governance framework, including enterprise database analysis and entity-relationship modelling for multi-country, multi-agency data systems. Evidence ↗
2002–20111 contribution
United Nations CGIAR
- Developed a digital crop-identification platform combining taxonomies, scientific datasets and AI-assisted recognition for agricultural biodiversity research.
Profile16 contributions
Across the career
- AI, Data and Technology Executive with 15+ years of international experience in artificial intelligence, Generative AI, machine learning, data strategy and enterprise transformation, combining executive-level AI strategy with hands-on technical leadership and principal engineering experience in AI and Generative AI.
- Builds and directs hybrid teams of people and AI agents, including swarms of AI agents, drawing on 30,000+ prompts of hands-on generative AI work.
- Specialises in Generative AI, LLMs, RAG, Agentic AI, predictive analytics, machine learning, MLOps, LLMOps, AI governance, AI factories and model evaluation; designs enterprise AI and data strategies and defines AI technology stacks and long-term machine-learning roadmaps through hands-on technical leadership.
- Designs enterprise AI governance frameworks covering use-case intake, risk classification, data provenance, human oversight, evaluation, deployment approval, monitoring and retirement.
- Combines AI product management, solution architecture, model evaluation, question design and rubric development.
- Translates emerging AI capabilities into practical business operations, including document ingestion, segmentation, forecasting, optimisation, personalisation and conversational assistance.
- Specialises in data products and AI strategies for pharmaceutical, biotech, clinical-development and biomedical environments.
- Leads enterprise AI literacy and adoption through training, prompt libraries, self-service resources, communities of practice and departmental champions at Gavi and EPAM.
- Translates ambitious strategy and complex technical possibilities into practical operating models, technology and delivery roadmaps, business cases, investment priorities, partnerships and measurable delivery outcomes.
- Converts unstructured scientific or commercial information — menus, catalogues, images, PDFs — into structured operational data using computer vision, LLMs, extraction pipelines and governed data models.
- Designs agentic development methodologies in which autonomous AI agents support architecture, code generation, testing, security review, documentation and delivery governance, along with hands-on agentic AI architectures and platforms.
- Applies continuous code generation, architecture review, security scanning, test-driven development and human oversight to AI-native engineering.
- Applies Generative AI to merchant operations, customer engagement, catalogue management, forecasting, personalisation, route optimisation and fraud detection.
- Designs natural-language business assistants and conversational customer experiences using multi-LLM orchestration.
- Uses predictive RFM segmentation, demand forecasting, personalisation and analytics to support customer retention and commercial performance.
- Long-standing advocate of FAIR principles and responsible AI, covering governance, risk and compliance (GRC), ethics, GDPR, the EU AI Act and FINMA requirements.
Skills9 contributions
Skills & applied knowledge
- AI and Generative-AI technologies: GANs, LLMs, agentic AI, FinOps, Trustworthy AI, MLOps, LLMOps, AI Factory strategy, MCP, A2A, multi-agent systems and Azure AI Foundry.
- Enterprise productivity, automation and security: Microsoft 365, Project, Visio, Power Automate, Copilot, Entra, Azure Key Vault, Defender for Cloud, Sentinel, Azure Policy and Purview.
- Has used AI development tools since 2014, with experience in complex AI-assisted and agentic software development: Cursor, Claude Code, ChatGPT Codex, Kimi, DeepSeek, Z GLM, n8n, Obsidian, LangGraph, Midjourney, Seedance 2.5, PyTorch, Google Gemini and multi-agent workflows.
- MLOps operating models, deployment practices, monitoring, automation and lifecycle governance.
- Artificial intelligence, data mining, neural networks, reinforcement learning, CNNs, LSTMs, attention mechanisms, Bi-GRUs, data engineering and data federation.
- Knowledge-management and semantic-AI methods: taxonomy, ontology, graph neural networks, semantic analysis, sentiment analysis and natural-language processing.
- Strong communication and presentation skills: technical evangelist with international keynotes and publications, communicating complex AI, data, investment and risk topics clearly to executive committees, boards and senior stakeholders.
- Five years of hands-on experience with Large Language Models and LLM-based applications.
- Model Context Protocol (MCP) and Generative Engine Optimization (GEO): making products and content usable and discoverable by AI assistants and agents.