Tools I've built with Claude to explore what responsible AI can do in public
services.
Building solutions is half of the work. Getting institutions to adopt change
is the other half, and what I've spent my career on.
01Streamlining service delivery
Clear Letter
government notices in plain English
Rewrites the dense notices agencies send for benefits delivery (determinations, renewals, document requests) into plain language while keeping the legal substance intact.
Relevance
Solving government's communication problem at the source. A resident who understands a notice can act on it.
Tools
Claude · RAG · Python NLP · React · Python/FastAPI
Status
Prototype
Scholarship eligibility quiz
find what you qualify for in 3 minutes
12 questions over a rules engine mapping to ~40 state workforce programs. A proof of concept designed with a state workforce agency so residents can quickly find programs they qualify for.
Relevance
Cuts through dozens of confusing programs to show people the ones that fit.
Status
Prototype
02Informing leaders & decisions
AI benchmarks dashboard
the pace of AI, at a glance
Tracks the tests used to measure AI against human ability (coding, reasoning, science, and more) and shows how fast the gap is closing.
Relevance
Gives public-sector leaders a clear-eyed read on how fast this is moving.
Tools
Claude · live benchmark data
Status
Live (internal)
Curated intelligence brief
a daily brief on AI for public services
Scrapes ~80 sources, summarizes what’s new, and pushes a morning brief. Started as mine; now keeps other teammates current too.
Relevance
Keeps overloaded leaders current without adding to the noise.
Tools
Cron · Claude · RSS
Status
Live (internal)
Custom course-to-audio tool
tailored AI learning leaders can take anywhere
Turns AI curriculum into an audio course, custom-built to what a leader needs to learn, and publishes it to the podcast apps they already use. Leaders can finish it on a commute.
Relevance
Helps busy public-sector leaders build AI fluency on their own terms.
Tools
Claude · ElevenLabs API
Status
Live (internal)
03Enabling others to build & create
Research OS
Civilla's research practice, rebuilt AI-native
A research system built in Claude Code: a reusable plugin of skills and agents over version-controlled repos that sets up a new study with one command. It runs every step from raw interview to clean transcripts, interview profiles, and synthesis across multiple frameworks.
Relevance
Lets a small team do rigorous, traceable qualitative research at a scale that usually takes a much larger one.
Tools
Claude · GitHub · Granola · Slack
Status
Live (internal)
Fieldwork
AI-moderated research interviewer
Built to run real interviews at scale, asking probing follow-ups and clarifying questions, then returning synthesized findings and every transcript. Not a replacement for in-person research, but a way to extend its reach.
Relevance
Lets institutions hear from far more residents and frontline staff.
Tools
Claude
Status
Prototype
Deep Dive
a full research study on demand
A 6-agent workflow that takes a public-sector leader’s question and runs the desk research for them: scoping the problem, scanning the existing evidence, and synthesizing it into a study they can act on. The result shows what’s already known and where to focus next.
Relevance
Puts rigorous desk research within reach of any leader.
Tools
OpenClaw · Figma · Telegram
Status
Live
Claude Code × Paper
brand-compliant design without a designer
A connection between Claude Code and Paper that lets staff with no design background generate fast, brand-compliant outputs. My team uses it daily.
Relevance
Puts professional, on-brand output in reach of non-designers, a real public-sector constraint.