I build agentic AI systems that automate real-world workflows.

Automating supply chain processes using agentic AI and computer vision as part of the AI Ops Integration team.

• Partnering with internal supply chain stakeholders to identify and automate manual operational processes using agentic AI and computer vision.
• Diving into research on applying agentic AI and computer vision techniques to supply chain problem spaces.

Designed, developed and deployed 3 AI-powered products with total ROI of £250,000 automating back-office workflows (internal audit) as part of the AI Lab in a finance environment.

• Automated the quarterly governance workflow for audit committee report generation using LLM summarisation, contextual numerical data, and automated report generation.
• Constructed a risk and control matrix builder using pgvector, tsvector, RRF and LLM reranking, covering risk identification from audit info, control mapping and extraction from large policies, control-to-risk validation, and Excel export.
• Built a concurrent batch evaluation pipeline assessing 246 enterprise policy documents against auditor-defined governance rubrics, with S3 PDF ingestion, structured LLM evaluation, confidence-based risk classification (green/amber/red), and automated remediation feedback — parallelised with retry handling for Bedrock throttling.

Designed an agentic Text-to-SQL pipeline to reduce financial reporting latency.

• Implementing a production-grade, agentic AI Text-to-SQL pipeline, reducing financial reporting latency by automating queries across 100M+ rows, cutting analyst workflow time by 85%.
• Selected as a success-story highlight in the director-level Monthly Business Review (MBR), giving the project VP-level visibility.
• Presented project outcomes to senior finance leadership, including Amazon Finance's AI Transformation Leader (within 3 reporting layers of Amazon CEO Andy Jassy).

Led a team of 6 to construct the MVP of our web application to reduce food waste.

• Developed a web application for three different user flows involving Stripe payment.
• Pitched to accelerators, clients and external stakeholders.
• Landed Barclays Eagle Labs partnership and multiple clients.

Built a proof-of-concept passenger-facing web application with real-time messaging via WebSockets, an Azure pipeline, and airline system integration.

• Designed an event-driven architecture bridging on-prem authentication with Azure IAM.
• Engineered an ETL pipeline to streamline billing calculations for the Azure pipeline.

Led a team of 7 to construct the MVP of our mobile application for our two-sided marketplace.

• Developed the Golang API, WebSockets-based notification service, and React Native UI.
• Constructed a Next.js web application, presenting our company's values and processes for user acquisition.
• Landed 2 MFI partnerships.

Constructed an app to help HPV patients through access to diagnostics and professional consultants.

• Developed the UI using React Native within a team of 8.
• Set up CI/CD pipelines using Github Actions and policies for pull request approvals encouraging shared responsibility for production code.
• Balanced work alongside two other startups.

Engineered the backend of V2 of the mobile application while tackling critical MVP tickets.

• Engineered backend APIs in Node.js/Express for mobile CRUD operations, designing PostgreSQL schemas and DynamoDB tables with global secondary indexes to optimize query performance and retrieval speed.
• Implemented a Socket.IO-based messaging system with reconnection logic, enabling real-time chat and active-user tracking.
• Migrated session tracking from PostgreSQL to Redis with TTL-based caching, reducing user lookup latency by 90% and ensuring scalable real-time updates

Collaborated in an artificial intelligence project that plans to improve educational services where students' attention profiles are monitored and shared with parents through their progress reports.

• Built an eye-tracking and blink-rate monitoring pipeline and triggered screenshots upon >5s absence to support progress reports.
• Experimented with Deep Learning models in Keras for eye-tracking and applied data augmentation to expand a limited dataset, prevent underfitting and improve generalisation.