Hours - AI Researcher
An AI research assistant that remembers. Hours learns your context over time, your projects, preferences, and research history, delivering personalised insights that get smarter with every conversation.
Hours' memory capabilities are powered by a complex RAG system architected by Sam and built by George and JK. The infrastructure uses vector databases to store and retrieve user context semantically. Conversations, research findings, preferences, and project details are chunked, embedded, and indexed for instant retrieval.
The technical challenges included designing chunking strategies that maintain coherent context across long conversations, optimising embeddings for both speed and accuracy, building query systems that surface relevant memories without hallucinating, and scaling the system to handle growing context windows as users accumulate history. The platform retrieves personalised context in milliseconds, making every interaction feel like a continuation rather than starting fresh.
The architecture supports horizontal scaling by design, meaning growth is solved by provisioning infrastructure rather than refactoring code. Load balancing distributes queries across multiple search nodes, database replicas handle read traffic in parallel, and the embedding pipeline processes requests asynchronously to prevent bottlenecks. The hard architectural decisions were made upfront, so scaling became an operational task, not an engineering challenge.
Get in touch
I'm always interested in exploring new opportunities, collaborating, or exchanging ideas with like-minded individuals. Feel free to book a call or email me if you'd like to see my portfolio deck or to discuss a potential project.