RouteMind is self-improving agent memory on MongoDB Atlas: a car teaches itself a real San Francisco drive, from Shack 15 to Ocean Beach, by learning from its own wrong turns. At each junction it recalls which way to turn from a memory of over a thousand vectors in Atlas, most of them wrong. After every trip it scores which recalled memories actually helped, keeps the useful ones, and deletes the ones that mislead. Over a handful of trips the route locks in, with no training and no retraining, while the same memory with learning switched off never leaves the first block. The point is a kind of memory that plain retrieval cannot provide. RAG ranks by similarity and never learns from what happened next, so an agent repeats the same mistake forever. In RouteMind, every possible turn is a document in MongoDB Atlas holding a Voyage AI embedding and a usefulness score; recall is one Atlas Vector Search query, and the agent writes the outcome back to Atlas and prunes the misleading memories. Because every candidate turn at a junction scores the same vector similarity, similarity alone cannot choose, and the usefulness learned and stored in Atlas is what picks the right turn. You can try it on any route: type a start and destination and it pulls the real roads from OpenStreetMap, builds the memory live, and drives it. It is built on MongoDB Atlas Vector Search (memory and recall), Voyage AI (embeddings), Raindrop (observability), Modelence (the deployed dashboard), and ElevenLabs (narration). Live demo: https://tets-mv4eiaze9dc-prod.cloud.modelence.app
More from The Harness Engineering & Model Wrangling #SFTechWeek Hackathon
MongoMation is an autonomous PLC engineer. You describe a machine in plain English, and Claude agents write IEC 61131-3 PLC code, compile and deploy it to a…

LabRat is a self-improving harness for a physical lab agent, running on a real bench: two SO-101 arms, a 3018 CNC carrying a servo pipette I designed during…

Recursive Harness Scientist: an AI scientist that improves its own scientific method, and must replicate every gain on data it has never seen. Why it…

Super Bid is a self-improving AI estimator that teaches itself to price bids the way a specific company would. It practices on the company’s real past bids…

A robot that gets better at handling objects it’s never seen, without retraining. After every round, an AI coach reviews what broke and rewrites the robot’s…
WORLDSTATE gives physical spaces a memory. Any phone becomes an observer: open a link, point the camera, and WORLDSTATE tracks what's inside fridges, shelves…
