Ravindu Pahasara
DevOps / DevSecOps / Cloud · Independent AI Systems R&D
DevOps, DevSecOps, and cloud engineering alongside independent R&D in persistent and governed AI systems.
Research
- TH-01: Memory & Continuity — How do we distinguish durable experiential synthesis from generative hallucination?
- TH-02: Governed Autonomy — How to guarantee policy invariants survive arbitrary planning outputs?
- TH-03: Verification & Telemetry — What verifiable telemetry constitutes proof of task completion?
- TH-04: Substrate-Agnostic Identity — What core invariants define agent continuity when model weights are replaced?
Writing
- Memory Is Not Truth — Why persistence is dangerous without verification, and why synthetic recall must be treated as hypothesis.
- Execution Is Not Success — Why zero exit codes tell us nothing about intent, and how verification differs from running tools.
- What Does Identity Mean When the Model Is Replaceable? — Substrate independence and the preservation of agency across generational foundation model upgrades.
Engineering work
- Multi-Region Cloud Platform & Autonomous Infrastructure — A reference design for a GitOps platform that reconciles declarative cluster topology, replaces unhealthy nodes automatically, and promotes releases through canary progressive delivery.
- SLSA-3 Verifiable Delivery & Zero-Trust Supply Chain — An engineering pattern for supply-chain security that blocks unsigned or un-attested code anywhere between pull request and production container start.
- Self-Mitigating Telemetry & High-Reliability Platform — A system study of an observability stack that combines eBPF kernel instrumentation with automated mitigation designed to act before human alert thresholds are reached.
- Persistent AI Runtime & Inference Backplane — An illustrative AI platform architecture supporting stateful multi-agent execution, sandboxed code execution, and semantic caching.