Robert – LLM, Python, Typescript

AI coding agents drift away from specifications as a project evolves. Robert designed and runs a production multi-agent system that breaks specifications into tasks and keeps agent output in line with the documentation.

• Designed, implemented, and ran cahoots.cc as a production multi-agent software planning system.
• Used Event Modeling to break down specifications into task structures.
• Implemented GitHub-based detection of drift from documentation.
• Designed specs-as-constraint guardrails.
• Generated TypeScript interfaces, GWT test skeletons, and architectural linter rules to keep agent output in sync with specifications.

In multi-agent systems, agents usually re-query for context at every step, which wastes tokens and adds latency. Robert built a context engine that keeps each agent's context current by pushing relevant updates to it as new data arrives.

• Developed Contex as a context engine for multi-agent AI systems.
• Implemented hybrid BM25 and vector search.
• Built event-sourced PostgreSQL persistence.
• Delivered events via Redis pubsub and webhooks.

A client needed to find relevant records across millions of scraped pages. Robert owned the work end to end, from architecture to AWS deployment, and delivered a release-ready semi-automated discovery pipeline with hybrid search.

• Developed a semi-automated web scraping and discovery pipeline.
• Created hybrid vector and keyword search using OpenSearch kNN/BM25 and OpenAI embeddings.
• Handled architecture, implementation, Cloudflare bot-auth integration, and AWS deployment end-to-end.
• Delivered the solution ready for release to the client team.

Enterprises deploying AI agents need to see and stop risky agent behaviour in real time. As a founding engineer, Robert built the evaluation side of a platform that monitors agent actions and blocks unsafe requests.

• Built evaluators for AI agents running in various frameworks.
• Built a pipeline that consumed Traceloop and OpenAI traces and ran evaluations on LLM inputs and outputs.
• Led a major backend refactoring that increased throughput by 60% and improved developer experience.

An American sports technology company is developing an online platform for sports betting and casino gaming. Users can bet on major sports, play in an online casino, and receive bonuses and rewards within a single ecosystem.

• Designed and implemented a real-time loyalty earnings calculation service handling 1,000+ requests per second at 2ms p99 via gRPC.
• Used a parameterized payout function instead of a key-value lookup table with session-based caching.
• Built and supported a GraphQL API for mutations and real-time updates of betting product state.
• Developed monitoring and debugging tools for the Kafka broker and consumers.
• Instrumented the service using Datadog, OpenTelemetry OTLP, and tracing.
• Wrote the department’s Java unit-testing standards and practices document.

A SaaS platform for visual teamwork and collaboration, allowing teams to brainstorm, plan, create diagrams, and work together on a shared digital whiteboard. The platform also uses AI features, templates, and integrations to streamline team processes and decision-making.

• Reduced database read-writes by 50% by implementing CQRS architecture and event-driven APIs with Kafka and PostgreSQL.
• Built a high-performance event dispatcher using Redis pub/sub and streams.
• Led the refactor of the company API from pure JavaScript to TypeScript.
• Developed integrations including Figma, PowerPoint, Power Automate, and Azure DevOps.
• Worked on the core services team creating the data access layer and refactoring old code.

A company in the mortgage/financial services sector that provides technology and operational solutions for lenders, mortgage servicers, and investors. The company uses its proprietary EXOS technology platform to automate and support processes related to mortgage origination and servicing, property valuation, title/closing, and default services.

• Designed and built a greenfield backend in C# .NET Core on Azure.
• Implemented Terraform and Helm across 200+ microservices.
• Built a real-time virtual avatar system with a Python/FastAPI backend, Azure Speech Synthesis, and a Xamarin Android front end.
• Created prototypes and proofs of concept for infrastructure, architecture, AR, and machine learning.
• Contributed to the EXOS Inspect application during planning, exploratory, and implementation phases.