VanGuard

We built Vanguard to explore how persistent agent memory can help AI understand competitor strategies over time.

A few things we learned:

• Flat RAG has limitations: Top-k semantic search finds similar information, but can miss long-term cause-and-effect relationships.

• Memory changes the picture: Instead of looking at a single price change, Vanguard connects historical events to understand the bigger strategic shift.

• Hybrid retrieval works better: We combine semantic similarity, keyword matching, and temporal relevance to retrieve more meaningful context.

The key takeaway:

More context doesn’t always mean better reasoning. Structured memory matters.

Built with Go, Python, FastAPI, Hindsight, Next.js, TypeScript, and Three.js.

AIAgents #AI #Hindsight #AgentMemory #LLM #RAG #ArtificialIntelligence #SoftwareEngineering