Independent systems builder
Building visible, auditable AI systems.
I’m Ron Reed, founder of Ebysslabs. I build AI infrastructure focused on retrieval governance,
observability, auditability, and operational trust.
My current work centers on RISWIS, a governance layer for RAG and AI systems,
and rag-radar, an ecosystem signal tracker for recurring retrieval failures in production AI discussions.
Core thesis: Most AI systems don’t fail at generation. They fail at what they decide to trust.
RISWIS
Operational startup layer — retrieval governance before generation.
What it is
- Retrieval Integrity & Structured Weighted Information System.
- A governance layer that sits between retrieval and generation.
- Controls which retrieved sources are allowed to influence downstream LLM responses.
- Makes ranking decisions visible, auditable, and policy-controlled before context reaches the model.
Why it matters
- Semantic similarity can surface the wrong source.
- Stale, low-trust, or policy-conflicting documents can still look relevant.
- RISWIS separates semantic ranking from governance weighting.
- Semantic winner ≠ policy winner.
Links
rag-radar
Ecosystem intelligence layer — retrieval failure signals from the field.
What it is
- A lightweight signal tracker for recurring RAG and retrieval reliability problems.
- Scans real AI infrastructure discussions for production pain patterns.
- Tracks recurring issues like citation failures, reranker instability, stale retrieval, and trust conflicts.
Why it matters
- AI builders often over-optimize demos while production reliability breaks later.
- rag-radar helps identify where retrieval systems are actually failing in the wild.
- Those signals help shape RISWIS and future governance tooling.
“The system works in demos but breaks in production.”
Current output
- Weekly signal reports.
- Retrieval reliability observations.
- AI infrastructure commentary.
- Production failure pattern tracking.
Supporting systems
CAS 2.0
Long-horizon drift & stability — observed, not steered.
What it is
- A methodological evaluation system for long-horizon observation of adaptive dynamics.
- Frozen parameters by construction; no optimization, tuning, retraining, or corrective control.
- Evaluated across 525+ structured configurations and 25.5+ million simulation steps.
Artifacts
FARMSLite
Local-only lifestyle journaling — privacy-first by design.
What it is
- An offline-first lifestyle journaling app designed around reflection and local ownership.
- No accounts. No tracking. No analytics. No cloud dependency.
- Designed as a quiet mirror for reflection, not a coaching or advisory system.
Links
Operator stance
Creator, builder, founder, infrastructure voice.
Focus
- AI governance infrastructure.
- Retrieval observability.
- Auditable decision systems.
- Operational reliability for AI systems.
- Privacy-first software design.
Ethical stance
- Safety: visible decision paths over hidden control.
- Privacy: local-first design where possible.
- Presence: systems built to support reflection, not extraction.