Skip to content
Open to interesting problemsLondon, UK

Andrei
Neagoe.

Full-Stack Software Engineer

Cloud & Infrastructure · Python Back-ends · React Front-ends

I build the infrastructure and the interface on top of it. At Framestore I design and ship serverless AWS services and the Terraform that provisions them, and maintain the event-driven messaging platform connecting the company's internal applications.

Alongside that I am the sole engineer behind Garzoni, an AI-powered financial learning platform live on web, iOS and Android with 150+ users. I care about the numbers that follow a decision — latency, cost, reliability.

View work→neagoeandrei23@gmail.comCV
Andrei Neagoe at his Queen Mary University of London graduation
andrei_neagoeLDN · GMT+1
Role
Associate Software Developer
Company
Framestore
Since
Jul 2025
Building
Garzoni
Based
London, UK
01 / Work

Products I own end to end.

Two things I built and still run — one a shipped consumer product on three platforms, one a studio site. Both are live, and both are mine from schema to pixel.

Garzoni

Live on web, iOS and Android

AI-powered financial learning platform

Founder & sole engineer2024 — Present

A personal finance education platform built the way a language app is built: ten-minute lessons, daily streaks, spaced repetition, and AI tools that let you rehearse the decision before you make it. I built and shipped the entire product solo — API, web app, mobile app, billing, infrastructure and content pipeline.

lines across the monorepo
~250klines across the monorepo
commits over 21 months
900+commits over 21 months
shipped platforms
3shipped platforms
users on web, iOS and Android
150+users on web, iOS and Android

Architecture

A pnpm monorepo spanning a Django 5.2 / DRF / Celery API, a React 19 web app and an Expo mobile app sharing a single TypeScript core — one set of types, one set of design tokens, three surfaces.

  • Django 5.2
  • DRF
  • Celery
  • React 19
  • Expo
  • TypeScript
  • pnpm

AI layer

An AI tutor built on OpenAI function calling with RAG over the curriculum (pgvector), voice tutoring through Whisper and TTS, receipt analysis via GPT-4o vision, and a learning path re-ranked daily by an LLM — hash-short-circuited so unchanged inputs never pay for a second inference.

  • OpenAI
  • Function calling
  • RAG
  • pgvector
  • Whisper
  • GPT-4o vision

Commercial layer

Three-tier subscriptions with cross-platform entitlements through RevenueCat — Stripe Web Billing, App Store and Play IAP reconciled into one source of truth — plus Plaid open-banking budgeting with statement import, and per-tier quotas and paywalls.

  • RevenueCat
  • Stripe
  • App Store IAP
  • Play IAP
  • Plaid

Platform

Gamification (streaks, daily and weekly missions, leagues, leaderboards, spaced-repetition review), JWT auth with Google and Apple sign-in, brute-force protection, reCAPTCHA Enterprise, Sentry, and Celery scheduled jobs behind 20+ backend test suites.

  • JWT
  • OAuth
  • reCAPTCHA Enterprise
  • Sentry
  • Celery

Production

Run end to end: Docker, Railway, Vercel and EAS store releases, Cloudflare edge caching, health-gated deploys and automated backups, with CI enforcing pinned dependencies, vulnerability auditing and secret scanning — plus EN/RO localisation across web and mobile.

  • Docker
  • Railway
  • Vercel
  • EAS
  • Cloudflare
  • CI/CD
Read the case studygarzoni.appApp StoreGoogle Play

Monevo

Live

Design and build studio site

Design & build2025 — Present

A studio site for the client work I take on outside Framestore. Built as a single long-form page with a WebGL fluid-simulation hero rendered on a real-time canvas, motion tuned to stay under a frame budget, and typography doing most of the work.

real-time fluid sim hero
WebGLreal-time fluid sim hero
live client projects shipped
4live client projects shipped

Front-end

React and Vite with Motion for choreography, a hand-rolled OGL fluid simulation behind the hero, and a token-driven light/dark system that flips without a flash.

  • React
  • Vite
  • Motion
  • OGL
  • Tailwind

Delivery

Hand-built rather than page-built: semantic markup, structured data and metadata, and Core Web Vitals treated as an acceptance criterion rather than a post-launch cleanup.

  • SEO
  • Core Web Vitals
  • Accessibility
monevo.tech
02 / Experience

Where the production work happens.

Serverless services, the Terraform under them, the message bus everything else talks through — and the ten-odd internal products that sit on top.

Framestore

Associate Software Developer

Jul 2025 — Present

London

My team owns the internal applications production runs on — planning, video review, file transfer, task import — so the work spans the whole surface: a new serverless AWS service and the Terraform behind it, the event-driven bus every other application talks through, and the React front-ends people actually sit in front of. Around ten products in a year.

Repeat link loads17×
348ms348ms

ElastiCache layer in front of RDS Postgres

Public Links

Average response11×
3.30s3.30s

Memory profiling, a 1MB package and a scheduled warmer

Public Links

Cold invocations33×
100%100%

From every request paying the start-up cost to almost none

Public Links

  • Public Links — serverless back-end

    Built the service end to end: API Gateway in front of a Step Functions Express chain across five Python Lambdas over RDS Postgres, batch S3 pre-signed URLs, and an ElastiCache layer that cut repeat link loads from 348ms to 21ms. The cache key had to be agreed with the legacy application, which invalidated on a different key — a rename stayed invisible until the TTL expired.

    • API Gateway
    • Step Functions
    • Lambda
    • RDS Postgres
    • ElastiCache
    • S3

  • Infrastructure as code

    Provisioned the full stack in Terraform — Lambda, API Gateway, CloudFront, WAF, ElastiCache, VPC, IAM — with per-environment remote state and GitLab CI deployment, including isolated per-developer AWS environments now standard on the team.

    • Terraform
    • CloudFront
    • WAF
    • VPC
    • IAM
    • GitLab CI

  • Hardening and cold-start work

    Locked the public endpoint behind AWS WAF and a CloudFront origin lock, then took cold invocations from 100% to 2–3% and average response from 3.3s to 290ms through Lambda memory profiling, an 18MB → 1MB deployment package and a scheduled warmer. A single logging standard across the chain means one query follows a request from edge to response.

    • AWS WAF
    • CloudFront
    • Performance
    • Profiling
    • Observability

  • Events Bus

    Maintained Framestore's messaging platform on RabbitMQ, MongoDB and Ansible: proved with data that long message processing was starving broker heartbeats, moved callbacks onto a thread pool and acknowledged after processing rather than before, added per-service broker authentication, and led the client rollout across 12 downstream applications.

    • RabbitMQ
    • MongoDB
    • Ansible
    • Concurrency

  • Features owned end to end

    Impersonation sessions in Filestore — permission model, nested session, a permanent indicator in the UI, chained impersonation blocked, a separate audit trail per session, and the support documentation. In Fetch, product features and a back-end refactor: domain-specific exceptions, a monolithic module split per entity, and a health check that checks something.

    • Django
    • Python
    • Auth
    • Auditing
    • Refactoring

  • Front-end and real-time

    Multi-participant remote cursors and canvas annotation over shared video in Framecast using WebRTC and Janus DataChannels, and Flock's React scheduling Gantt — grouping counters, an availability heatmap, and date-windowed fetching so a wide range no longer stalls the grid.

    • React
    • WebRTC
    • Janus
    • Canvas

A year across the estate

  1. Jun — Aug 2026

    Public LinksNew public service

    A serverless AWS service replacing the path clients use to view and download their material, taken from a query audit through to observability — and the React UI wired to the real API, loading states included.

  2. May — Jul 2026

    Events BusInternal messaging

    Heartbeat starvation on long-running tasks, a routing bug on identifiers containing dots, an authentication rework separating native broker users from hashed keys for REST and UI, and alerting into Sentry and Google Chat.

  3. Mar — Jul 2026

    Legacy modernisationInternal services

    Python 2 → 3 and 3.8 → 3.9, RPM packaging → Docker and Ansible, CentOS 7 → EL9. Plus a read-only proxy in front of the Jira API — GET only, bodies dropped, cookies isolated, rate limited, tokens masked in logs — and a webhook proof of concept driving an orchestrator workflow.

  4. Jan — Apr 2026

    FetchTask import for production

    Rebranded end to end — code, URLs, logo, log dashboards — then shipped prep imports, department exclusions, template autocompletion and filtered item counters, over a back end split into per-entity modules.

  5. Nov 2025 — Feb 2026

    FilestoreFile storage

    Impersonation sessions built from the permission model up to the audit trail, alongside two repositories consolidated, a database migration, and the staging and production deploys behind them.

  6. Aug 2025 — Jan 2026

    FlockResource planning

    A React Gantt UI: counts at every grouping level, a fast toggle for on-hold bookings, an availability heatmap from spike to implementation, and requests chunked by time window.

  7. Jul — Sep 2025

    FramecastCollaborative video review

    React over WebRTC and Janus: remote cursors for every participant, annotation on a canvas layer above playing video, group-gated session controls, and favourite rooms kept locally.

Brixton Radio

Data Automation Developer

Sep 2024 — Mar 2025

London

Designed the automated data pipeline behind a live broadcast schedule and its social analytics.

less manual data processing
40%less manual data processing
weekly shows managed
50+weekly shows managed
better metadata accuracy
30%better metadata accuracy

  • Automated data extraction

    Reduced manual data processing by 40% by designing an API-driven extraction system handling high-volume social media and analytics data.

    • Python
    • REST APIs
    • Automation

  • Scheduling integration

    Integrated Google Calendar and Forms to streamline management of 50+ weekly radio shows, improving metadata accuracy by 30% in a fast-moving broadcast environment.

    • Google APIs
    • Ops tooling
03 / Stack

What I reach for.

Weighted towards what I actually ship with — AWS and Terraform on the infrastructure side, Django and FastAPI behind the API, React and Expo in front of it.

Languages

  • Python
  • TypeScript
  • JavaScript
  • SQL

Cloud & Infrastructure

  • AWS Lambda
  • Step Functions
  • API Gateway
  • CloudFront
  • AWS WAF
  • S3
  • RDS
  • ElastiCache
  • VPC
  • IAM
  • Amplify
  • Terraform
  • Docker
  • Ansible
  • GitLab CI
  • GitHub Actions
  • GCP

Back-end

  • Django
  • DRF
  • FastAPI
  • REST API design
  • Celery
  • PostgreSQL
  • MySQL
  • MongoDB
  • RabbitMQ
  • Redis

Front-end

  • React
  • React Native / Expo
  • Vue
  • Vite
  • Tailwind
  • HTML5
  • CSS3

Also

  • Git
  • WebRTC
  • LLM APIs (OpenAI, Claude)
  • Sentry
  • ELK
  • Figma
  • Agile / Scrum

Spoken

  • EnglishProficient
  • RomanianNative
04 / Background

Where it came from.

A BCS-accredited computer science degree, ongoing internal engineering programmes at Framestore, and the university work that led here.

Education

BSc (Hons) Computer Science

Queen Mary University of London

2022 — 2025

BCS-accredited degree from the School of Electronic Engineering and Computer Science.

Focused on software engineering, databases, web development and neural networks, with team projects run to a full delivery cycle from requirements through to demo.

  • Procedural & Object-Oriented Programming
  • Algorithms & Data Structures
  • Computer Systems & Networks
  • Database Systems
  • Software Engineering
  • Web Programming
  • Neural Networks & Deep Learning

Earlier work

Training & awards

  • ML at Framestore

    Framestore · internal training

    Four-day programme covering ML theory and hands-on model building, MLOps development and deployment, data provenance and asset tracking in production, diffusion models and large language models.

  • Web Development Group

    Framestore · internal training

    Ongoing 30+ session programme on the company's standardised web stack: React, Vue, Django, FastAPI, REST API design and lifecycle, Docker, GitLab CI deployment, load balancers, Google SSO/OAuth, ELK logging and GCP cloud workshops.

  • Royal Academy of Engineering masterclasses

    Graduate Engineering Engagement Programme · Nov 2023 — Jun 2024

    Selected for an eight-month series on professional networking, public speaking and personal branding, alongside sessions on engineering career pathways, sustainability, and taking research through to commercialisation.

  • National Olympiad accolades

    Mathematics & Computer Science

    Placed in multiple national high-school olympiads in Romania.

05 / Contact

Got something worth building?

Email is the fastest way to reach me — infrastructure problems, AI product work, or a back-end that has outgrown its architecture. I read everything.

neagoeandrei23@gmail.com