3 ms·
I just wonder how do you filter out signal from noise - does it self correct wrong memories?
by imhoguy 3mo ago
I just wonder how do you filter out signal from noise - does it self correct wrong memories?
- vshulcz 3mo ago[dead]
- latchattack 3mo agoI agree that similarity search can't surface reliable project judgment alone. How do you know what was true, what was rejected, and what's obsolete since changes in the project? I am actually building Open-Latch for this reason. It leverages project decisions, facts, lifecycle state, reconciliation, enforcing cited evidence and a judgement gate before coding agents act. GitHub: https://github.com/open-latch/latch https://github.com/open-latch/latch I felt the same annoyance as OP mentioned, but focused on agents violating project judgement and direction. Obviously we are optimizing for some different things than deja: deja uses no-LLMs and exact verbatim transcript retrieval. Latch uses LLMs, ranking, etc, to emphasize strong judgment, correction, and enforcement over dynamic project flow.