Work

Selected work

Engagements Giorgi has owned or led in production. Outcomes below are from public descriptions of that work — not invented testimonials.

Let’s talk about your product

METIS / Lloyd’s MGA · AI Engineer / Fractional technical lead

Operational compliance platform for a Lloyd’s MGA

A Lloyd’s managing general agent ran its regulated work — senior-manager accountability, product approval, evidence, and follow-up actions — across spreadsheets, ad-hoc forms, and what individual people happened to remember. Nothing enforced who had signed off on what, so every audit and every handover started again from the beginning.

METIS is the system of record for that governance now. Approvals move through defined steps, evidence and templates sit in one place, owners are notified, and the platform decides whether a product is approved rather than a spreadsheet. Regulated status is recorded as work happens, so an audit can be answered from the system.

How it was built
What was built
A Next.js and Supabase application covering SMCR, Product Approval Process, principle-level maturity tracking, broker management, evidence libraries, action items, and in-app and email notifications. Postgres, Row Level Security, RPCs, triggers, and Edge Functions keep workflows transactional and access controlled, while a unified document and template model supports PAP obligations, Fair Value, Consumer Duty, Target Market, and maturity assessments.
Technical challenges
  • Encoded the multi-step PAP workflow so product status is derived from completed steps rather than manually edited, while remaining race-safe under concurrent approvals.
  • Unified assessments and obligations in one document engine without breaking existing workflows, permissions, seeds, or templates.
  • Enforced Row Level Security, grants, and GraphQL access so clients cannot bypass business rules, routing privileged operations through RPCs.
  • Built event-driven notifications from domain events through an outbox to email, with environment guards that prevent non-production environments from messaging real users.
  • Kept production and development seed overlays, migrations, and Edge Function secrets aligned so deployments remained operable.
What shipped
  • Shipped a production-ready governance platform spanning SMCR, PAP, maturity, brokers, evidence, and actions.
  • Consolidated fragmented assessment surfaces into a reusable template and document architecture.
  • Hardened product and document creation so critical launch artefacts, including the concept note, cannot silently fail.
  • Stabilised Edge Function service authentication and notification delivery, including email allowlisting for safe non-production use.
  • Established migration-first documentation, incremental seeding, CI, and role-based Playwright coverage as part of the delivery pipeline.
Stack
Next.js · TypeScript · Supabase · PostgreSQL · Row Level Security · Edge Functions

Flighter Group · Full-stack / fractional technical lead

Aviation onboarding and document intelligence

Flighter onboarded aviation applicants over email and shared drives. Identity checks, certificates, employment history, compliance review, signatures, and approvals were all tracked by hand, evidence was scattered, and there was no dependable record of who had reviewed or approved what.

Onboarding runs as one workflow. Documents are collected and read automatically, anything the system is unsure about goes to a person, signatures and approvals are captured in order, and every decision leaves a trail. Turnaround shortened and the repetitive checking largely went away — without removing the human approval gates a regulated business needs.

How it was built
What was built
A production onboarding and certification platform with secure authentication, encrypted document storage, AI-assisted extraction and validation, event-driven processing, human approval gates, and third-party electronic signatures. I owned the product architecture, frontend, backend, cloud infrastructure, document-processing workflows, integrations, deployment pipeline, and technical delivery.
Technical challenges
  • Extracted and validated structured information from varied identity, employment, and aviation documents while routing uncertain results through human review.
  • Designed resilient, idempotent workflows that could recover safely from retries, partial failures, and concurrent updates without skipping a compliance step.
  • Protected sensitive applicant data through layered access controls, encryption, tenant isolation, and audit trails aligned with SOC 2 readiness.
  • Integrated a third-party electronic signature service into asynchronous document and approval lifecycles, including status reconciliation and failure recovery.
  • Kept frontend, backend, cloud infrastructure, environments, and deployment automation aligned as the product and its regulated workflows evolved.
What shipped
  • Shipped the complete product from architecture through production across applicant onboarding, document collection, compliance review, signatures, and approval.
  • Reduced onboarding turnaround and manual checking by automating document extraction, validation, reminders, and workflow progression.
  • Replaced fragmented records with consistent certification data, centralised evidence, and an auditable history of decisions and actions.
  • Combined automation with explicit human approval gates so operational efficiency did not weaken compliance controls.
  • Established repeatable cloud infrastructure and deployment practices that supported ongoing product development and operational reliability.
Stack
TypeScript · React · AWS · Serverless Architecture · Infrastructure as Code · AI Document Processing

Eolas Medical · Senior full-stack engineer

Hospital knowledge app with AI search

Clinicians needed one trustworthy place for clinical guidelines, their own hospital’s documents, and onboarding material — on a ward, on a phone, and without exposing anything a given user should not see.

The app is in production with hospitals and individual clinicians, web and mobile running on the same backend, and AI-assisted search across millions of guidelines that stays inside each user’s access rights.

How it was built
What was built
High-availability hospital application with content management, secure messaging, fine-grained access, and AI-assisted search across millions of medical guidelines.
Technical / AI challenge
HIPAA-sensitive data, cross-platform sync, and an AI search path that stays inside access controls.
Stack
React Native · React · AWS · GraphQL · OpenAI

Phoenix Court / LocalGlobe · Senior full-stack engineer

AI analytics for investment teams

Investment analysts were spending too long turning raw market signals into something a partner could act on.

Reported 30% higher engagement on insights, 40% less analysis time, and 60% faster deploys after the pipeline work. Trends and funding opportunities reach the team automatically, with the reasoning visible instead of a black box nobody trusts.

How it was built
What was built
React and Node features plus Python/PostgreSQL tools that detect trends and surface funding opportunities, with an AWS pipeline to ship them.
Technical / AI challenge
Useful AI for domain experts — predictive analytics and workflow automation without a black-box that nobody trusts.
Stack
React · Node.js · Python · OpenAI · AWS