Sourav Sheth

Knowing what changed and what actually matters across a codebase is a different problem from just reading diffs, and I don't think existing code review tools really solve it. The "what matters" part is doing the heavy lifting in that tagline. How does RepoLens decide something is worth surfacing versus noise?

Document automation sounds simple, and then it gets incredibly complicated the moment you hit real edge cases. conditional clauses, dynamic tables, and multi party signing flows. Curious how Anvil handles those without the user having to write logic by hand

The reusable skills part is what caught my attention. That's where this either becomes something teams build around or just another assistant they use a few times and forget.

I keep seeing "semantic layer" come up here but I'd want to understand what that means concretely. Are we talking about a layer that exposes annotated schema metadata to the model, or something closer to NL to SQL? The failure mode I keep running into with agent-first backends is hallucinated foreign keys and migrations that look fine until you run them.

Spent some time going through it and the structure already feels better. It is less “wall of text” and more guided, which was badly needed still a lot of concepts to wrap your head around, but at least the entry point isn’t as intimidating now

I have been manually cleaning up things like build folders and caches for a while, mostly because I never fully trusted automated tools to not remove something important. This seems like it is trying to bridge that gap by adding more visibility instead of just automation. Curious how well it handles edge cases though, especially when files are tied to active projects or environments.