Praetor

AI | Agents | Governance | LLM Infra

Praetor is a governance-first AI orchestrator for legacy-to-cloud migration in regulated enterprise environments. The reasoning pipeline is decomposed into isolated agent stages (assessment, design and visualisation), each with typed handoffs, human-in-the-loop gates, and auditable outputs. A single-model chain provides neither the traceability nor the review points a regulated migration needs.

Built on open-weight models (Gemma 4, GPT-OSS) rather than closed APIs so the full chain stays inside the trust boundary, with retrieval on Weaviate and evaluation via DeepEval.

  • Python
  • LangChain
  • Weaviate
  • DeepEval
  • Gemma 4 (E4B, 26B)
  • GPT-OSS 120B

Automation & Intelligence Layer

Automation | AI | LLM Infra | Backend

Custom internal tooling for a boutique services firm, composed of a Python monorepo of research pipelines and a composable skill library that accelerates back-office operations.

The skill library follows a microservices pattern: atomic reference skills composed by task skills, so each dimension of an output is independently versionable and regression-tested via an Expected / Defect / Undocumented grading framework. Direct prompting doesn't scale for domain-specific work where consistency drifts without a testable surface. Research pipelines transform raw external data into structured, ranked signals suitable for downstream decisions. Operational data lives in a Notion layer of eight linked databases, modelled relationally.

  • Python
  • Anthropic API
  • Claude Code
  • Notion
  • Git

Editorial Site & Design System

Design | Development | Design System | Motion

An editorial site for a luxury lifestyle brand, featuring a bespoke Astro build with a design system and page motion, built to read as an editorial property rather than a service page.

Astro over Next.js: the site is content-heavy and interaction-light, so SSR was overhead the site didn't need, and shipping minimal JS mattered more than framework ergonomics. Colour runs a role-token layer over palette tokens, so seasonal cadence shifts happen as atomic edits rather than a repo-wide sweep. Page transitions use a bespoke GSAP curtain over Astro's default cross-fade, with the logo drawn onto the curtain via drawSVG, timed to the lift.

  • Astro
  • TypeScript
  • Tailwind CSS v4
  • GSAP
  • Lenis
  • Astro ClientRouter
  • Figma

Event-Driven Deployment System

Backend | Cloud | Distributed Systems | Security

Distributed cybersecurity operations run in environments with unreliable network coverage. The platform had to keep working under constrained conditions and stay auditable for government users.

Built on event-driven, serverless AWS primitives (Lambda, DynamoDB, CloudFront) for fault isolation and predictable performance across loosely coupled services. Applied domain-driven design and standards-based identity (OAuth2, Cognito) so downstream teams weren't paying the cognitive cost of ad-hoc APIs. Adaptive caching, capacity tuning and event orchestration cut throughput and latency by 40%. Seven-engineer team, with responsibility across backend services, identity workflows, and observability.

  • JavaScript
  • TypeScript
  • React
  • AWS
  • Node

Health Platform

Backend | Database | Data Governance | Public Sector

A national public-sector workforce of 1M+ employees was running employee health and wellbeing on fragmented manual processes, resulting in data inaccuracies, delays, and compliance risk.

Oracle APEX integrated with AWS Workspaces, rather than a pure-code build: low-code was faster where the UI didn't need to be bespoke, and custom services picked up the gaps where governance couldn't be met by defaults. Data governance was the primary constraint. Encryption, RBAC, audit trails, and least-privilege throughout cut production vulnerabilities by 30%. Custom single-click export replaced cross-departmental manual pulls, taking 50% off manual effort and 45% off response times.

  • Oracle APEX
  • PL/SQL
  • SQL
  • AWS
  • JavaScript
  • Node