Skip to content
AB
All projects

Project

Live, with users

EuclidPRO.ai

Secura Eyes Informatics · 2024 – present

EuclidPRO.ai logo

Custom software platform built on skeleton_foundation: digital transformation, audited AI agents and applied intelligence. Its Field Operations module is live and in use by an HSE team of 35 users across 350+ locations.

  • Fastify
  • React
  • PostgreSQL
  • PgBouncer
  • Row-level security
  • Self-hosted maps
View PRDVisit live siteGitHubDiscuss this work

Purpose

EuclidPRO.ai builds custom software for organisations on one proven foundation, skeleton_foundation. Each engagement maps how the work really flows, assembles the modules, roles, scoring rules and limits, and delivers under the client's own address, logo and colours, with records on euclidpro infrastructure or in a database and storage the client already owns.

My role

Head of AI, Product & Technology at Secura Eyes: product owner, architect, UI/UX lead, full-stack builder and go-to-market lead. Built the first version solo; now leads a team of three.

Highlights

  • Three tiers: digital transformation, audited AI agents and applied intelligence
  • Field Operations module live and in use by an HSE team: 35 users across 350+ locations
  • Every site on a self-hosted map, coloured by risk and by how long since anyone looked
  • Operator console with capability switches per plan or organisation; every change versioned and audited
  • Each organisation gets its own address, logo, colours and roles

Three tiers

  • Digital transformation: forms, spreadsheets and email chains become governed workflows, with the data in one place and every step recorded
  • AI agents: triage submissions, draft reports and answer questions about an organisation's own data, acting only within a person's role, with every action audited
  • Applied intelligence: risk prediction, anomaly detection and document understanding, trained on the client's data and delivered inside the tools people already use

Field Operations (Geolocation)

  • Every site on a self-hosted map, coloured by risk and by how long since anyone looked
  • Next visit decided automatically: never-visited sites first, then the longest since a visit, tie-broken by risk band and distance from base
  • Outward and return routes on the road network from each site's base; the round trip is credited when the visit is completed, then frozen
  • No third-party map or routing calls
  • Sites scored by an audited model: weights, thresholds and override rules edited visually, tested in a what-if simulator and published as audited versions
  • Field app installs to the home screen; every form autosaves on the device and sends itself when the network returns; inspectors see the sites closest to them
  • Coverage and risk mix per base, distance reports per inspector exported to CSV, and a one-click site report card
  • Sites imported from CSV or GeoJSON: columns map themselves and every row is previewed on a map before saving

Capabilities

  • Operator console: guided onboarding for organisations, plans, and capabilities switched on or off per plan or organisation in seconds
  • Every console change versioned and audited; organisations' own admins can change the settings marked as theirs
  • Per-organisation address, logo and colours; records kept apart; roles defined by each organisation; access follows role
  • Invitations, and delegation that expires on its own
  • Individual accounts for people outside an organisation, with their own plan, personal projects and shared groups
  • Bring-your-own database and storage, verified before use
  • Deployment options: SaaS, private cloud, on-premise, air-gapped and Docker

Governance & quality

  • Access checked on every request
  • An administrative record that can't be edited afterwards
  • Passkeys for administrators, extra checks before sensitive actions and protection against automated attempts
  • AI agents act only within a person's role, and every agent action is audited
  • Encrypted backups and tested restores; availability, response times and backup schedules agreed per client in an SLA
  • Tenant data isolated by schema and enforced row-level security
  • Human approval kept in the loop before anything is published

Architecture & delivery

  • Runs on skeleton_foundation: Fastify + React, PostgreSQL 16 + PgBouncer
  • Single Linux server install, with optional Cloudflare Tunnel
  • Product features added as modules: tables, routes and screens

Current stage & roadmap

  • Current stage: Field Operations module live and in use
  • Next: Task module
  • Next: Approval module
  • Offered tiers: AI agents and applied intelligence

Outcomes

  • Field Operations module live and in use by an HSE team: 35 users, 350+ locations
  • Public product site live at euclidpro.ai

Skills demonstrated

  • AI product strategy
  • Multi-tenant SaaS architecture
  • Full-stack development
  • Self-hosted mapping and routing
  • Audited risk scoring
  • Linux deployment
  • Human-in-the-loop governance
  • UI/UX
  • Model evaluation
  • Enterprise GTM positioning
  • Team leadership

History · 2024 – 2026

Earlier version: generative AI training platform

The first EuclidPRO.ai was an AI-powered training and enterprise-intelligence platform. It turned minimal user inputs into complete training content and downloadable PDF presentations, with human control over every generated step. It started as training and compliance content generation for a 9,000+ multilingual shift workforce, moved to compliance and contracts, and then pivoted to today's modular platform.

What it did

  • Generated complete training content and downloadable PDF presentations
  • Multi-stage orchestration across course, module, slide, interaction and assessment planning
  • Mandatory human review and approval at every orchestration phase
  • Localisation, mobile-first learning and structured assessment orchestration
  • Mobile-first, multilingual, RTL/LTR and accessibility-aware design
  • Google Vertex AI and Gemini integration for structured generation
  • Working deployment on a Linux VDS

Governance & quality

  • Mandatory human approval gates at every orchestration phase
  • Human-in-the-loop publication control
  • Structured validation for accuracy, completeness and formatting
  • Responsible AI and auditability-oriented design choices
  • Evaluation against hallucination risk and output quality rubrics

Outcomes

  • Live hosted product on a Linux VDS with working training-generation capability
  • Demonstrated applied AI orchestration with strong governance controls

Earlier roadmap

  • Learner registration, assignments, records, dashboards, reporting, audit visibility and adoption tracking
  • Document intelligence, OCR ingestion, chunking, RAG and enterprise search
  • Policy, SOP, compliance and workflow automation modules

This roadmap was superseded by the 2026 pivot to the modular platform. Current stage: the Field Operations (Geolocation) module is live and in use, and the Task and Approval modules are next.