LIVE ON KANSAS CITY'S ACTUAL MUNICIPAL CODE
The Clarity Infrastructure for Government & Civic AI
Before you deploy AI to your constituents, ensure your source data is safe, structured, and clear. aclara.ai operates underneath existing systems to make complex municipal code and regulations human-verified and machine-ready.
Verified for humans. Structured for AI. Measured all the way down.
Meet the gatekeeper. Every unit below passes through it before it's ever served to a constituent.
ingestion pipeline
RUNS ONCE, BELOW YOUR STACK
The Unstructured Reality
$32B+ in non-defense AI deployment is currently gated by toxic, unstructured input data.
Government AI struggles at the input layer: difficult source content, contradictory information, siloed systems. Standard LLM wrappers inherit all of it.
✕ HALLUCINATED ADVICE
Wrappers answer anyway
NYC's MyCity chatbot answered questions beyond its source data — and told business owners to break the law. Raw legal text plus a generic LLM is a liability engine.
✕ TOXIC INPUT DATA
Raw code is unreadable — for humans and models
Municipal ordinances run on cross-references, repealed chapters, and grade-16 sentences. Retrieval over that text misreads questions, ignores queries, or shuts down entirely.
✕ STALE LAW
The corpus rots silently
Fees drift out of date, cited chapters get repealed, discretionary standards get auto-evaluated. No one is watching the inputs — until a constituent acts on a wrong answer.
The aclara Clarity Layer
How every tool runs better on us.
aclara sits between raw regulations and your customer-facing AI — delivering 100% deterministic, human-verified, structured data units. Nothing is generated at query time.
ch. 27 · fences and walls · Municode export
RAISE™ score → plain rewrite → expert verify → cite
chatbots · 311 · search · agents — every tool runs better on us
PASS units → served with citation · FLAG units → expert review · GAP → honest refusal
The data moat
The Power of RAISE™
aclara is calibrated against years of human expert-scored data pairs using the RAISE™ rubric — 5 dimensions, 35 standard items, operationalizing ISO 24495-1, the international plain language standard. It measures whether readers can find information, understand it, and act on it. Every routing decision in the pipeline depends on the score.
R
Relevance
Is this the information the reader needs?
A
Accessibility
Can the reader find and reach it?
I
Intelligibility
Can the reader understand it?
S
Suitability
Does it fit the reader and the genre?
E
Efficacy
Can the reader act on it?
Calibrated, not vibes.
0%
Alignment
On LLM-judged criteria, the engine lands within a single rubric level of a human expert 100% of the time. Machine checks match expert scoring exactly.
0%
First-pass approval
High-scoring units clear the auto-approve threshold with zero expert minutes. The remaining 22% route to human review — expert time goes exactly where the score sends it.
0×
Reader comprehension
Measured on real agency content: comprehension doubles and user errors fall by two-thirds after the clarity pipeline.
0 → 0.0
Penalty points / 1,000 words
A failing Public Works letter reworked to passing — with MQM-style, ISO-aligned quality measurement proving the move.
Independent validation
In NIH RADx evaluations, RAISE™ aligned perfectly with legacy instruments like PEMAT and SAM — while yielding far more actionable diagnostic output. Three independent instruments agree on direction. The engine's one known bias is conservative: it understates improvement. Every number we show is a floor.
Live case study · Kansas City, MO
Three moments. Zero improvisation.
This is aclara running on real law — Kansas City's actual fence-and-wall ordinance (KCMO Code ch. 27), ingested from the city's official code. Nothing below was generated at query time.
resident@kcmo:~$ Can I build a 6-foot fence in my front yard?▋
✓ ACLARA-VERIFIED · AUTO-APPROVED BY CALIBRATED RAISE™ ENGINE
On a residential lot, your fence or wall can be up to 6 feet tall. Some parts of your yard have lower limits — check the front yard and corner lot rules. Want to go taller? You can ask the Board of Zoning Adjustment for a special exception.
KCMO Code § 27-10(a) · Fences and walls · cited character-for-character
"(a) Fences and walls, generally. No fence or wall over six feet high shall be erected on a residential lot, provided that this height restriction is further limited in subsections (b), (c), (d) and (e) of this section."
reading grade 9.8 → 4.3 · deterministic rule evaluated · human-reviewed · zero expert minutes spent
Wifi off. Fully local. This is what an accountable civic AI layer looks like.
Strategy
Infrastructure, not a wrapper.
01
Productize
The clarity pipeline as recurring software: agency platform license plus usage pricing per document processed and verified. Land with a pilot, expand to platform.
~$850K already contracted across 4 agencies
02
Own the Model
Twenty years of expert-scored data pairs burned into our own fine-tuned weights — scoring and rewriting on-prem, no frontier API in the loop.
deployable air-gapped · no external calls
03
Clear Procurement
Enterprise security built for the public sector: SOC 2 Type II compliance track for FedRAMP standards, localized PII scrubbing, automated audit logging.
designed for the RFP, not against it
—
The moat, in the order it compounds
The RAISE™ rubric and the only calibration gold data for it — years of expert-scored government documents. The verification workflow governments actually require. Our own weights. Better frontier models make our ingestion cheaper; they don't replace the accountability layer.
authored by Romina Marazzato Sparano — director, MQM Council · contributor, ISO 24495-1
Founder & CEO
RominaMarazzato Sparano
The person who helped write the standard.
Romina Marazzato Sparano is the Founder and CEO of aclara.ai, the clarity data infrastructure for civic and enterprise AI. A leading global plain-language pioneer, Romina is a key contributor to the ISO 24495-1 international plain language standard, a director at the MQM Council, and President of Plain Language International Inc. (Plainlii). Over a 20-year career in translation, localization, and technical writing — including founding Plainlii and establishing the Translation & Localization Management MA program at MIIS — she has built the world's most robust repository of expert-scored language data. Through aclara.ai, she is productizing this deep data moat with the proprietary RAISE™ rubric, building the clear and secure pipelines required to make public sector AI safe, accessible, and trustworthy.
Enterprise / Gov-Ready
Built for Government Security & Sovereign Data
Air-Gapped Local Deployment
Zero external API calls. The full pipeline — scoring, rewriting, verification, serving — runs entirely inside secure agency firewalls. Your legal corpus never leaves the building.
wifi off · fully local · frontier vendors structurally can't offer this
Low-Compute Runs
A complete pipeline validation run costs less than $500 of compute. No GPU cluster, no per-token metering surprises, no cloud dependency in the budget line.
< $500 / full validation run
Procurement-Friendly
Actively designed for SOC 2 Type II and federal data-privacy compliance, with localized PII scrubbing and automated audit logging. Built for the sign-off chain a city attorney will actually sign.
SOC 2 Type II track · FedRAMP-aligned · audit logs by default
Your AI deployment is only as good as its inputs.
Start with an infrastructure audit: we score your source content with RAISE™, map the gaps, and show you exactly what your constituent-facing AI would inherit today.
- → Scored RAISE™ report on your real documents
- → Gap map: what your AI would inherit today
- → Response within 2 business days
Meeting at AGI Summit 2026? Let's talk before you leave — romina@aclara.ai