Eight years across backend platforms, data systems and AI infrastructure

Platform work for MuleSoft's AI products. Agent orchestration, the gateway in front of the model providers, and the governance that has to hold once real customers are pointing traffic at it.

• Multi-agent orchestration

  Contributed to the orchestration capability that lets a customer chain several specialised agents together inside a governed workflow, instead of pushing every task through one general-purpose agent.

• Governed LLM gateway

  Contributed to an enterprise LLM gateway that centralises model access, provider routing, usage governance and policy enforcement across multiple AI providers.

• Model Context Protocol policies

  Worked on the platform policies covering Model Context Protocol traffic between agents, tools and enterprise services: authentication, authorisation, context propagation and keeping tool access safe.

• Agent-to-agent communication

  Worked on policies and infrastructure supporting secure, governed communication between autonomous agents: discovery, identity propagation and authorisation challenges through controlled gateways.

• Usage and cost governance

  Built governance capabilities around virtual model credentials, enabling centralised access control, attribution and policy enforcement without exposing underlying provider keys.

• Multi-dimensional rate limiting

  Developing rate-limiting and cost controls for enterprise LLM usage across users, agents, models, providers, requests, tokens and budgets.

• Centralised access revocation

  Built the capability that lets an organisation cut off a specific user or agent from downstream AI resources straight away, evaluated as part of the gateway policy chain.

• Configurable request-key extraction

  Added configurable request-key extraction for gateway policies, so identity and quota keys can be derived from request context instead of a single hard-coded field.

• Agentic AI infrastructure
• LLM gateway
• MCP & A2A policies
• AI governance
• Distributed rate limiting
• Streaming & SSE

Data movement and privacy-aware processing for internal support systems, plus the backend services around them.

• Near-real-time data movement

  Led development of a data-movement capability that improved near-real-time availability of support information by removing a major dependency from the critical path.

• GDPR-aligned scrubbing pipeline

  Architected a near-real-time data-scrubbing system for support information, improving privacy enforcement without sacrificing operational availability.

• Support-case collaboration

  Built a case-collaboration capability connecting internal support systems, reducing friction in support-case resolution.

• Scalable web services

  Contributed to backend services with an emphasis on reliability, maintainability, performance and production readiness.

• Data pipelines
• GDPR-aligned processing
• Distributed services
• Reliability

Workflow tooling aimed at people who are not engineers, and a long-running effort to hold the product to real accessibility standards.

• Flow Designer

  Developed workflow-automation capabilities enabling non-specialist users to create and configure enterprise workflows more easily.

• Review and Submit

  Designed and implemented a review experience for enterprise requests in Workspace, helping users validate information before submission and reduce errors.

• Accessibility remediation

  Improved accessibility across enterprise workflow interfaces by resolving WCAG 2.0 compliance defects.

• Workflow automation
• Low-code experiences
• Accessibility (WCAG 2.0)
• Enterprise software

Backend work on Zoho Notebook, mostly around making transfers reliable and downloads fast.

• Pause-and-resume uploads

  Built reliable pause-and-resume support for file uploads, allowing interrupted transfers to continue without data loss.

• Evernote migration

  Built a high-performance migration path from Evernote to Zoho Notebook that avoided conventional API round-trips, reducing migration time and friction.

• Download server

  Designed a dedicated download-serving capability that optimised request handling and reduced download time by approximately 30%.

• Prediction features

  Built prediction features using the machine-learning techniques of the day. Worth saying plainly: this was years before the current wave of generative AI and has little in common with the platform work I do now.

• Backend product engineering
• Transfer reliability
• Performance optimisation
• Data migration