Senior AI Developer
Consultancy Outfit (CO-UK)
• Loophole Larry (UK legal AI): Building the RAG pipeline for an AI case solver over UK law, where a user submits acase and the system generates a legal strategy. Own the data pipeline, knowledge-base indexing and retrievaloptimization for long-document queries: a single “query” is a case file running to hundreds of pages rather than ashort prompt, which rules out standard RAG retrieval patterns. Work in progress as more data is ingested.
• Data privacy: Implemented PII anonymization and redaction of case data with Microsoft Presidio to meet UK GDPRrequirements before documents enter the index.
• CossmoneyX and CurrencyGram (fintech): Designed and pitched to the CEO a 13-agent compliance system thatassesses readiness for FCA registration (crypto-asset for CossmoneyX, e-money for CurrencyGram): scrapes theregulator’s published requirements, indexes them alongside the company’s own policies in a vector database, andruns 11 parallel analyst agents (AML, sanctions, Travel Rule, financial promotions, etc.) plus a coordinator and areviewer agent to produce a decision-ready readiness report with strengths, gaps and suggested improvements.
• Property Peak (UK mortgage platform): Scoped and prioritised nine AI agents (case tracking, summarisation,handoff checks, ticket triage, onboarding, commission updates) across six user roles, each with role-based dataaccess and GDPR constraints; building them as LangGraph agents behind FastAPI REST APIs with platform dataexposed via the Model Context Protocol (MCP).
Associate Team Lead - AI
Devsinc
October 2025 - April 2026
• Voice AI: Led the design and development of a real-time conversational video/voice AI agent with Speechmatics andDeepgram speech-to-text, RAG for grounded answers and LangGraph agents for multi-step dialogue.
• Latency: Migrated the stack from a Pipecat telephony prototype to the ElevenLabs API as the product moved tovideo, bringing end-to-end response latency to a consistent 950 ms p95; replaced a custom LangGraph web-searchtool with OpenAI’s built-in web search, which was simpler to integrate and faster.
• Deployment: Deployed the platform on AWS and set up production monitoring.
• Mentoring: Trained and mentored junior and intern data scientists on LLM application development, code reviewand deployment practices.
Math / Algorithms Expert
Zigron
Solar Power Generation and Optimization - Terrasmart (US)
• Sole developer and architect of an end-to-end engine that models a solar site and optimizes its energy production;reported directly to the Director.
• Built a 3D mathematical model of site layout, sun position over discrete time steps and inter-row shading betweenpanels, then optimized panel tilt angles to maximize total daily energy output.
• Solved the constrained optimization with MATLAB Global Optimization Toolbox solvers (pattern search, particleswarm, genetic algorithm) and tuned their parameters to speed up convergence.
• Reduced runtime by vectorizing the pipeline, parallelizing it with the Parallel Computing Toolbox and porting hotpaths to C++.
AI Dentistry - Automated Analysis of Dental X-rays
• Sole ML developer (initially); demonstrated the viability of AI for tooth numbering, dental diagnosis and bone-lossestimation from radiographs.
• Processed and annotated a dataset of several hundred dental images; achieved 90% accuracy on tooth numbering.
• Ran R&D on model architectures including DETR (DEtection TRansformer) and on preprocessing such as CLAHEand wavelet transforms to improve detection on low-contrast X-rays.
• Deployed the solution on a local server and presented the model’s capabilities and limitations to academics atUniversity College Dublin.
