Sentinel-E was inspired by a simple problem: most AI systems sound confident even when their reasoning is weak or hallucinated. I wanted to build a system where models could challenge each other, verify claims, and expose reasoning instead of acting like isolated black boxes. The idea evolved into a multi-model cognitive engine where agreement doesn’t automatically mean truth. That led to debate mode, evidence verification, confidence calibration, and transparency-focused reasoning. One of the biggest challenges was authentication and persistent memory architecture — especially handling cross-session recollection, user session continuity, JWT/auth synchronization, and reliable retrieval of user-specific data from the database without breaking isolation or context integrity. Managing orchestration between multiple models while keeping latency, session state, and memory retrieval stable was also a major engineering challenge.

sentinel-e

An AI reasoning engine built on disagreement

An AI reasoning engine built on disagreement