Hill Patel
AI Systems Engineer | Agent Infrastructure | AI Platform Arc
New to Contra
Hill is ready for their next project!
I spent the last year building the part of AI systems nobody puts in the demo: what happens when things go wrong.
Most agent frameworks are built to answer one question, what can this model do?
Nexus is built to answer a different one, what happens when it's wrong, and how do we know?
Here's what that actually means, in plain terms:
Every decision has one owner. No step in the system can quietly act without something else checking it. Writing, deleting, or changing anything real requires a human to say yes or not as a courtesy, but as a hard rule the system can't route around.
Nothing is remembered by guesswork. Every action is logged as it happens, in order, permanently. If something breaks, I don't have to reconstruct what happened from memory or hope the logs are enough. I can replay the exact sequence of events and see precisely where it went wrong.
Recovery isn't a footnote. A system that fails silently and reports success anyway is worse than one that crashes loudly. Nexus is built so failure is visible, traceable, and recoverable not swept under a "task complete."
The result feels less like an agent framework and more like an operating system, one designed around a simple belief: the hard problem in AI isn't making a good decision. It's knowing, with certainty, what actually happened after you made it.
Nexus v2.0 just shipped. It's open source, and it's the clearest version yet of everything above.
GitHub: https://github.com/STiFLeR7/nexus
(https://github.com/STiFLeR7/nexus)Would genuinely love feedback from anyone building production AI systems.
1
12
agentic-rag is a production-grade Agentic Retrieval-Augmented Generation system engineered to run reliably on consumer laptops (specifically RTX 3050 / 6GB VRAM). It rejects "black box" abstractions in favor of a controllable, graph-based architecture that prioritizes reasonability, reliability, and explicit failure handling.
0
10
Recoverable long-horizon AI agents, a framework-agnostic reference harness + recovery-faithful live benchmark. Thesis: "Checkpoints Are Compactions" via Re-grounding Recovery. 0.x: v1.0 held until a powered live-LLM study confirms the claims.
0
14
Governed AI execution platform with runtime orchestration, approval workflows, operational intelligence, recovery systems, and sandboxed autonomous execution.
0
33