Work
Projects
Production tools, research utilities, and infrastructure automation — built at the intersection of Lloyd's market expertise and modern AI.
Lloyd's Market News Digest
AI-powered daily intelligence for the Lloyd's insurance market. Monitors 50+ sources overnight, extracts the signal from the noise, and delivers a concise, structured digest before the market opens — with zero manual curation.
30 Days 30 Apps Challenge
30 AI applications for Lloyd's and specialty insurance, built and shipped in 30 consecutive days — 14 March to 12 April 2026. Underwriting, exposure management, claims, broking, compliance, leadership, and more.
Other Projects
Memex — Personal Second Brain
Production AI-maintained digital garden and Second Brain at memex.poovi.me — inspired by Andrej Karpathy's personal AI infrastructure work and Vannevar Bush's original memex paper.
Karpathy's vision: an AI that accumulates knowledge the way a human does. Here it's real — raw sources (articles, PDFs, transcripts) become an interlinked wiki maintained by Gemini. OpenClaw agents query it via MCP and publish findings to LinkedIn, all without manual curation.
Full-stack: Quartz static wiki deployed to Vercel, FastAPI automation engine with cron-triggered watcher, Gemini 2.5 Pro/Flash synthesis pipeline, Qdrant vector search (7,483 pages indexed), FastMCP stdio server for agent integration.
Framesense — Composition Coach
In ProgressAI-powered photo composition guide for iPhone using Claude Vision API to give real-time framing feedback.
Built to explore Claude's Vision capabilities in a mobile context — and to solve a real problem: getting actionable composition feedback without a photography tutor.
Full Expo/React Native app: camera integration via react-native-vision-camera, Claude Vision prompting, NativeWind styling, Zustand state management.
Zero-Employee Company Lab
ActiveActively running and testing the emerging category of AI-native 'zero employee' company platforms — Paperclip, OpenClaw, and Hermes AI — on private homelab infrastructure.
The Lloyd's market is 3–5 years from AI-native operations. Hands-on production experience today means I can speak from architecture, not theory, when clients or leadership ask whether this is real. I'm the only account manager I know running a company OS on a home server.
Live on Proxmox homelab: Paperclip as company governance OS (role/policy management, audited task execution) running a structured company instance; OpenClaw routing 11 named agents (FURY, ORACLE, WAYNE, CYBORG, BANNER, STARK, XAVIER...) via Telegram and Discord; Hermes AI orchestration under active evaluation.
MyStockify
Stock analysis toolkit for evaluating investment opportunities.
Built to automate my own investment research — combines financial APIs, multiple valuation models, and data visualisation into a single workflow.
Full Python build: data ingestion, scoring engine, Jupyter output.
Homeserver-as-Code
Infrastructure-as-code setup for a fully self-hosted home server.
Wanted reproducible, version-controlled infrastructure that could be rebuilt from scratch in minutes — no more manual configuration drift.
Docker Compose stack, reverse proxy, automated backups, and monitoring.
Docker DB Backup Automation
Scheduled, containerised database backup with configurable retention.
Solved a real gap: most teams run databases in Docker but rely on fragile manual backup scripts. This handles scheduling, rotation, and alerting automatically.
Shell scripting, Docker integration, cron scheduling, retention logic.
Claims Pipeline
Data transformation pipeline for insurance bordereaux processing.
Bordereaux data arrives in inconsistent formats from coverholders. This pipeline ingests, validates, and standardises the records for downstream processing.
Python ETL pipeline, schema validation, pandas transformations.