Josh Barrass

Designed a modular plugin system for dynamically loading third-party widgets and pages, enabling safe extensibility and accelerating integration for client modules — analogous to SDK/package design patterns. Built a Vue.js frontend for an industrial PLC-style robotics platform, improving UI consistency and system uptime. Developed backend C++ services and modular API endpoints to support secure third-party integrations with predictable deployment behaviour. Collaborated directly with clients to scope, build, and ship custom platform modules end-to-end.

Architected an agentic AI analyst system powered by OpenAI with structured tool-calling: an orchestration loop executes up to 12 tool calls per message — querying player aggregates, tactical patterns, economy data, and positional heatmaps — before synthesising natural-language coaching responses streamed to the client via SSE. Designed and maintained 23+ tool definitions consumed by the LLM, implementing server-side execution, result serialisation, and context injection — directly analogous to MCP-style tool-calling patterns for agents. Built end-to-end inference and retrieval pipelines: a curated knowledge base with weighted full-text search (title×10, tags×5, content×1), pro-play reference matching, and context-aware retrieval injected into system prompts for grounded AI responses. Scaled the full-stack platform (Next.js 15, TypeScript, MongoDB) to 75,000+ monthly users with real-time leaderboards, subscription billing (Stripe webhooks), and tiered access control. Engineered a multi-stage data pipeline on Cloudflare Workers and R2: chunked uploads reassembled server-side, parsed by a custom Worker, then split across MongoDB collections with automatic downsampling to stay within BSON limits. Deployed containerised workloads on AWS (ECR/ECS) with Cloudflare for DDoS protection, edge caching, and object storage — delivering low-latency, secure infrastructure at scale. Implemented production hardening: SHA-256 idempotency deduplication, IP-based rate limiting, anti-bot systems, and retry pipelines with dead-letter fallback storage.

Redesigned the architecture of a Python simulation tool, applying modern engineering standards to achieve a 200× speed improvement — the tool contributed to infrastructure bids exceeding £1 billion. Built and deployed an engineering platform using AWS Infrastructure-as-Code, delivering scalable, reproducible cloud environments. Implemented security hardening for C++ applications, ensuring reliability across production deployments.