AI & Workflow Automation

AI-assisted actuarial delivery

A practitioner using AI on real actuarial work — not an AI vendor. Data pipelines, disclosure drafting, multi-domain retrieval, and workflow orchestration, designed for regulated insurance and finance environments.

Capabilities

What I can deliver

AI-assisted data pipelines

Intelligent ingestion of policy, claims, cash-flow, and finance-system data
Automated validation, reconciliation, and gap detection across source systems
Contextual enrichment — attaching definitions, regulatory references, and lineage to each record
Human-in-the-loop review gates at every material stage of the pipeline

Automated report generation

IFRS 17 disclosure drafting — quarter-end and year-end narrative from structured model output
Actuarial working papers with sensitivities, systematic allocation ratios, and AJE / NDIC commentary
Regulator-response drafting with citation-backed answers to technical queries
Management and board packs generated from the same underlying result set

RAG knowledge retrieval

Multi-domain vector knowledge bases — regulatory, methodology, implementation, internal memos
Contextual-prefix embedding and distillation templates for regulatory standards and insurer financials
Eval framework with MRR / NDCG@5 / Precision@5 scoring and regression detection per domain
Scoped retrieval so client, regulator, and methodology corpora stay cleanly segregated

Workflow orchestration

Claude Code with custom skills for reconciliation, month-end, and compliance-check workflows
MCP server integrations across inbox, calendar, repo, and browser for end-to-end automation
Scheduled task orchestration — market scans, portfolio review, regulatory monitoring, digest generation
Structured memory and audit trails so every automated step is traceable and reviewable

Capability showcases

Work I can demonstrate

AI client case studies are still being built. In the meantime, here is what I maintain, use daily, and have published publicly — anchored in systems I have built, not slides.

Platform · Built internally

RAVEN — multi-domain retrieval system

Six knowledge domains indexed in a vector store: IFRS 17, banking regulation, SAM, SAMA, plus working email and Slack corpora — 5,000+ vectors total.
Contextual-prefix embedding pipeline with distillation templates for regulatory standards and insurer financial statements.
Eval framework (MRR / NDCG@5 / Precision@5) with per-domain baselines and regression detection on retrieval quality.
3D corpus map with clustering, domain toggles, and reranker-backed retrieval on the front-end.

Demonstrates what a working actuarial knowledge platform looks like end-to-end — ingestion, chunking, retrieval, evaluation.

Practice · Daily use

A practitioner’s automation stack

Custom Claude Code skills for reconciliation, month-end analysis, budget review, and regulator-response drafting.
MCP server integrations across Slack, Gmail, Chrome, and GitHub — inbox triage, repo review, and browser tasks from one surface.
Tiered memory layer: session-load context, on-demand reference retrieval, and archived historical state.
Scheduled orchestration for market scanning, portfolio review, and weekly research digests.

I practise what I propose — the workflow an AI-assisted consulting practice actually runs on, not slides.

Publications

A point of view on AI in insurance

A survey of practitioner-grade applications across underwriting, claims, and actuarial work — read 7 ways insurers are using AI right now.
A framework for scoping, piloting, and operationalising AI in regulated insurance environments — read the AI use-case lifecycle.

Applied, governance-aware thinking — tied to what I actually see in regulated insurance work.

Tools & Stack

What I actually use

Claude Code

First choice

My first-choice surface for AI-assisted actuarial work. Custom skills, MCP server integrations, structured memory, and scheduled task orchestration — daily driver across reconciliation, reporting, and research workflows.

Anthropic API

Claude Sonnet as the default model, Opus for complex reasoning, Haiku for high-volume low-cost steps. Used directly for production pipelines where Claude Code is not the right surface.

OpenAI API

Embeddings for retrieval pipelines. Selective use for specific document-understanding tasks where the OpenAI model is the stronger fit.

Python + MCP ecosystem

Python for data pipelines, transformations, and eval harnesses. MCP servers for Slack, Gmail, GitHub, Chrome, and custom domain tools — assembled into end-to-end automation.

Vector databases

Pinecone as the primary store for RAVEN’s multi-domain corpora. Cloudflare Vectorize evaluated for edge-deployed retrieval where low-latency inference matters.

Cloudflare Workers

Edge deployment for retrieval APIs, webhooks, and always-on automation. Paired with Workers AI and AI Gateway for model routing and caching.

Delivery partner

Engagements on this page are delivered in partnership with Virtual Actuary. Virtual Actuary is the delivery vehicle; strategic direction, scoping, and technical leadership sits with me.

Start a conversation

Want to explore what AI can do in your team?

Pipelines, reporting automation, or a scoped retrieval platform — book a call to walk through where AI is the right fit in your workflow and where it is not.

Book a Call

Writing on this topic

ai-governance5 Apr 2026

The AI Use Case Lifecycle: Why Governance Is a Cycle, Not a Checklist

A 5-stage iterative lifecycle for AI use cases in insurance — from context and design through ongoing monitoring — and why treating governance as a one-off approval gate is where most programmes break.

Read →

ai-governance31 Mar 2026

Your GLM Is AI Now: How the MAS MindForge Definition Changes What Insurers Must Govern

Under MAS MindForge guidelines, any model that learns from data is classified as AI — including your pricing GLM. Here is what that means for governance.

Read →

ai-governance30 Mar 2026

7 Ways Insurers Are Using AI Right Now: Production Use Cases Across the Value Chain

From underwriting risk scoring to personalised pricing, here are 7 AI use cases already deployed in production at leading insurers and financial institutions.

Read →

Reference