AI·rete·RAG claims its released platform keeps authoritative verdicts in a deterministic Rete engine while using retrieval and LLMs only for grounded explanations, potentially making AI-assisted domain decisions repeatable and auditable.

state: seedheat: lowuncertainty: mediumconvergesscott: highauditable-rag rule-engines decision-supportAI·rete·RAGZahara Hussain

What is this?

ai·rete·rag is a decision-intelligence platform (launched via Show HN) that pairs a deterministic, pure-Python Rete rule engine with retrieval-augmented generation: YAML-defined rules fire on asserted facts to produce the verdict ('the what'), while retrieval over the user's own policy documents plus an LLM generates a plain-language, citation-grounded explanation of a decision that has already been made ('the why') — the LLM structurally cannot alter the outcome. The builder is cited in the case as Zahara Hussain; the supplied snippets don't independently confirm that name or other details about the person or the company. The site claims eight built-in domains, a rule graph with nested logic and forward chaining, salience-based conflict resolution, and a full decision audit trail.

Why it matters to Scott

AI·rete·RAG independently ships the exact architecture Scott's Decision Authority Infrastructure and 'Stop Asking AI Why It Decided' argue for: a deterministic rule engine owns the verdict, the LLM is structurally confined to citing its own policy docs for the explanation, and a full audit trail rides along — a public, dated third-party implementation of his doctrine and a ready-made case study, especially for where it falls short of the full DAI spec (no signed authority, proof-carrying proposals, or decision attestation packages, and the supply-snippets don't confirm the named builder). The radar already tracks sibling episodes (Elevarq, Railo) but not this one, so this is a fresh receipt, not a repeat.

ip:framework.decision-authority-infrastructureip:concept.deterministic-coreip:source.stop-asking-ai-why-it-decidedip:concept.policy-as-source-codeip:concept.explainability-trapradar:elevarq-postgres-llm-truth-boundaryradar:railo-deterministic-security-patching

queries asked of Scott's wikis

LLM-as-explainer vs LLM-as-decider — has Scott written about separating reasoning from explanation in agent pipelines?
deterministic tool/verifier layers around probabilistic models — agent harness patterns where code, not the model, holds authority
RAG grounded explanation and citation — Scott's RAG / knowledge-system designs for citing source documents
rule engines (Rete, Drools, YAML rules) vs LLM decision-making — past projects or positions
auditable AI decisions in regulated domains (lending, fraud, triage) — compliance and reliability arguments Scott has made
agent-maintained wikis: could structured rule/fact stores play the role ai·rete·rag gives documents?

Measured heat

now 0 pts/hpeak 0 pts/hcomments 0/hpeers p14momentum: steady1 platformsage 456h

points/hour across evidence · reading as of 2026-10-12 02:59:37.977291+11:00 · deterministic, not a model opinion

How the heat travelled

09-22 16:15	⭐ origin directly observed	Show HN: AI·rete·RAG – a Rete rule engine decides, RAG explains why ZaharaHussain on hacker news	—
09-22 16:15	amplified on hacker news 👑	hn.story.49803683 ZaharaHussain	peak 46 · 11 comments · 100% of case engagement
09-22 16:21	our radar first saw it · +0.1h	discovery anchor: hn.story.49803683	—

pace: p62 vs 1032 stories at the 336h mark (now 456h old) — ahead of openai-chatgpt-financial-services (1.0x), behind linum-jit-ddt-training-efficiency (0.9x)

Evidence (1) — ⭐ canonical anchor

source	object	author	score	comments
🟧 hn ⭐	Show HN: AI·rete·RAG – a Rete rule engine decides, RAG explains why Retrieved article excerpt Open article · Retrieved 2026-09-23T16:27:38.891789+00:00 ai·rete·rag [Platform](https://ai-rete-rag.com/platform)[Patterns](https://ai-rete-rag.com/patterns)[Docs](https://ai-rete-rag.com/docs-site)[Writing](https://ai-rete-rag.com/blog)[Pricing](https://ai-rete-rag.com/pricing) Sign in with Google Get started Decision intelligence # Deterministic decisions, told in plain language. ai·rete·rag pairs a Rete rule engine with retrieval-augmented generation. Rules decide the what — auditable and repeatable. Retrieval explains the why — grounded in your own documents. Start buildingSee it decide ↓ Try it — no signup ## See a real decision, live. These run against the live engine on real demo policies. Change the facts, run it, and read the deterministic verdict with a grounded explanation. Loan underwritingFraud screeningClinical vitals Drag the credit score down and watch the verdict flip from approved to declined. Credit score745 Debt-to-income0.31 Loan-to-value0.75 Run decision → ⚖️ Adjust the facts and run the decision to see the verdict and explanation. This is one of eight built-in domains. Bring your own rules and policies to build your own.Start free — no card Query incoming case → Rete rules the "what" ⇄ Retrieval the "why" → Decision + Explanation auditable answer Rule engines Precise at structured fact matching and deterministic logic — but unable to explain themselves in natural language. Language models Fluent at explanation and synthesis — but weak at exact, repeatable, auditable logic. ai·rete·rag runs them together. ## Three ways to wire rules and retrieval. All three run live today — rules scope retrieval, documents feed working memory, and every verdict is explained in plain language. Compose them per request. 01● Live Rules → Retrieval ### Rules as retrieval filters Rules narrow scope before retrieval — a cardiac case fetches only cardiology sources, for focused context and far less room to hallucinate. 02● Live Retrieval → Rules ### Retrieval into working memory Documents are parsed into facts — entities, dates, obligations — and asserted into the session. Rules then fire on what was read. 03● Live Decision → Narrative ### Post-hoc explanation The engine fires first and produces a decision trace. Retrieval is invoked only to generate a human-readable explanation, grounded in your source documents. ## Full audit trail, not a black box. Every decision links back to the exact rules that produced it — and the ones that almost did. Explore Policy Rules → Policy Rules ### Every rule, laid out Browse the full rule catalog per domain — conditions, salience, and verdicts — instead of hunting through YAML files. Decision audit ### See why, not just what Click into any decision and see exactly which rule fired — and why every other rule didn't, down to the value that missed the threshold. Conflict detection ### Conflicts caught automatically Static analysis flags when two rules with different verdicts could both match the same case, before it becomes a production surprise. ## Ready to wire up your first domain? Bring your rules and documents. ai·rete·rag handles the rest. Open workspace → ai·rete·rag [Platform](https://ai-rete-rag.com/platform)[Patterns](https://ai-rete-rag.com/patterns)[Docs](https://ai-rete-rag.com/docs-site)[Writing](https://ai-rete-rag.com/blog)[Pricing](https://ai-rete-rag.com/pricing)[Terms](https://ai-rete-rag.com/terms)[Privacy](https://ai-rete-rag.com/privacy) © 2025 ai·rete·rag	ZaharaHussain	46	11

source

object

author

score

comments

🟧 hn ⭐

Show HN: AI·rete·RAG – a Rete rule engine decides, RAG explains why

Retrieved article excerpt

Open article · Retrieved 2026-09-23T16:27:38.891789+00:00

ai·rete·rag

[Platform](https://ai-rete-rag.com/platform)[Patterns](https://ai-rete-rag.com/patterns)[Docs](https://ai-rete-rag.com/docs-site)[Writing](https://ai-rete-rag.com/blog)[Pricing](https://ai-rete-rag.com/pricing)

Get started

Decision intelligence

# Deterministic decisions, told in plain language.

ai·rete·rag pairs a Rete rule engine with retrieval-augmented generation. Rules decide the *what* — auditable and repeatable. Retrieval explains the *why* — grounded in your own documents.

Start buildingSee it decide ↓

Try it — no signup

## See a real decision, live.

These run against the live engine on real demo policies. Change the facts, run it, and read the deterministic verdict with a grounded explanation.

Loan underwritingFraud screeningClinical vitals

Drag the credit score down and watch the verdict flip from approved to declined.

Credit score745

Debt-to-income0.31

Loan-to-value0.75

Run decision →

⚖️

Adjust the facts and run the decision to see the verdict and explanation.

This is one of eight built-in domains. Bring your own rules and policies to build your own.Start free — no card

Query

incoming case

→

Rete rules

the "what"

⇄

Retrieval

the "why"

→

Decision + Explanation

auditable answer

Rule engines

Precise at structured fact matching and deterministic logic — but unable to explain themselves in natural language.

Language models

Fluent at explanation and synthesis — but weak at exact, repeatable, auditable logic. ai·rete·rag runs them together.

## Three ways to wire rules and retrieval.

All three run live today — rules scope retrieval, documents feed working memory, and every verdict is explained in plain language. Compose them per request.

01● Live

Rules → Retrieval

### Rules as retrieval filters

Rules narrow scope before retrieval — a cardiac case fetches only cardiology sources, for focused context and far less room to hallucinate.

02● Live

Retrieval → Rules

### Retrieval into working memory

Documents are parsed into facts — entities, dates, obligations — and asserted into the session. Rules then fire on what was read.

03● Live

Decision → Narrative

### Post-hoc explanation

The engine fires first and produces a decision trace. Retrieval is invoked only to generate a human-readable explanation, grounded in your source documents.

## Full audit trail, not a black box.

Every decision links back to the exact rules that produced it — and the ones that almost did.

Explore Policy Rules →

Policy Rules

### Every rule, laid out

Browse the full rule catalog per domain — conditions, salience, and verdicts — instead of hunting through YAML files.

Decision audit

### See why, not just what

Click into any decision and see exactly which rule fired — and why every other rule didn't, down to the value that missed the threshold.

Conflict detection

### Conflicts caught automatically

Static analysis flags when two rules with different verdicts could both match the same case, before it becomes a production surprise.

## Ready to wire up your first domain?

Bring your rules and documents. ai·rete·rag handles the rest.

Open workspace →

ai·rete·rag

ZaharaHussain

Interpretation history

2026-09-23T20:34:12Z
grounded: converges/high — AI·rete·RAG independently ships the exact architecture Scott's Decision Authority Infrastructure and 'Stop Asking AI Why It Decided' argue for: a deterministic
2026-09-23T04:22:56Z
case created — The usable product demonstrates a concrete architecture separating deterministic decisions from generated explanations.

Decision trace

10-01 16:29review_screenjev screen: no material development (noul=0.15)
09-26 00:30review_screenjev screen: no material development (noul=0.11)
09-24 06:34groundAI·rete·RAG independently ships the exact architecture Scott's Decision Authority Infrastructure and 'Stop Asking AI Why It Decided' argue for: a deterministic rule engine owns the verd
09-24 04:15createThe usable product demonstrates a concrete architecture separating deterministic decisions from generated explanations.