Ben Keilman
Implementation & Enablement

I help non-technical teams adopt new technology — and I build the tools myself.

Business operations analyst with nearly a decade inside a government agency turning dense technical systems into the training, documentation, and rollouts thousands of staff actually use. For the last three years I've built my own AI tools — RAG systems, autonomous agents, automation — on Claude and the Anthropic API. The rare combination: I can run the adoption and be a credible technical partner.

See the work My first 90 days → Get in touch

Boston, MA · in-office or hybrid preferred, open to remote · government & regulated-industry native

How I work

Adoption is the deliverable. The build makes it credible.

Most implementation work fails at adoption, not installation. I lead the human side — the trainings, guides, and change management that get people to actually use a tool — and because I build the same kinds of systems myself, the demos and playbooks come from someone who has shipped, not someone reading a script. See exactly how I'd run it in a new role →

Enablement & adoption

Trainings, road shows, job aids, train-the-trainer, and community-of-practice programs that drive new tools into daily use across hundreds of staff.

Build & deliver

Three years building RAG systems, agents, and automation on the Anthropic API — working pilots and freelance systems clients run in production.

Regulated-environment fluency

Nine years inside a government compliance environment — audit, data classification, access control — the instincts responsible-AI deployment demands.

Adoption track record

Change management, at real scale

At the Massachusetts Department of Transitional Assistance, getting people to adopt new tools and processes is the core of my role — not a side effect of it.

~800
staff I drive tool & process adoption across
1,200
page Online Guide I develop & maintain
~150
person office onboarded to the internal GenAI platform
9+ yrs
in a government compliance environment

Selected work

Tools I've built and put into real use

A mix of internal pilots inside a regulated agency and freelance systems clients run in production. The proprietary ones are shown through architecture and sanitized artifacts; the open ones link straight to the code.

Open source · live

claude-autofix — a self-healing error system

Python · Anthropic API · Claude Code

When a production process throws an error and no one's awake to fix it, this doesn't just retry — it reads its own source code, reasons about the failure, rewrites the code that caused it, restarts the process, and emails a report. Autonomously, inside hard guardrails (rate limits, a daily attempt cap, deduplication, and escalation to a human). Ran in production for 53 days on a personal data pipeline; this repo is the genericized, open extraction, with a no-API-key demo and real fix reports the agents wrote themselves.

53 active days in production 42 error types diagnosed 691 errors → 1 fix (one day) runnable in 10s, no API key
View on GitHub →
Flagship

Policy Navigator

Python · FastAPI · ChromaDB · Voyage AI · Anthropic API · HTMX

A multi-source RAG over 11,000+ regulatory sections across four authoritative sources — ask a plain-language policy question, get a cited, synthesized answer. Cross-checks sources to surface gaps and drafts new policy content from research.

View on GitHub →
Open source

agent_lab — a fleet of scheduled agents

Python · Claude Code

A framework for running specialized, scheduled Claude Code agents that share a workspace, message each other through files, and accumulate durable knowledge over time — extracted and genericized from a live multi-agent system.

View on GitHub →
Detail coming

Draft-Document Peer Reviewer

Python · Anthropic API · python-docx

An agentic reviewer that checks draft documents against department standards — accessibility, inclusivity, formatting, completeness — and returns structured feedback for human signoff. A worked before/after of AI removing toil while keeping a human in the loop.

Detail coming

Regulations Comparator

Python · Anthropic API · ChromaDB · Voyage AI

A 4-stage agentic pipeline (classify → triage → deep investigation → cross-check) detecting coverage gaps across 480+ regulatory sections before auditors find them — a generalizable pattern for comparing any two compliance bodies.

Skills

What I bring

Enablement & communication

Change & adoption management · training-program and curriculum design · large-scale documentation and playbooks · train-the-trainer and power-user programs · onboarding and stakeholder enablement · plain-language translation across audiences · accessibility (508 / WCAG) · UAT · presenting to leadership.

AI / LLM (hands-on practitioner)

Claude / Claude Code (3+ yrs) and the Anthropic API · RAG · agentic and multi-agent workflows · prompt engineering · tool-calling orchestration · MCP · embeddings (Voyage AI) · agent safety guardrails · model and vendor evaluation.

Responsible AI & governance

Responsible-AI guidance for non-technical staff · government compliance environment (9+ yrs) · data classification, audit, and access-control discipline · proactive gap detection and audit-readiness.

Engineering & platform fluency

Python · FastAPI · SQL / SQLite · ChromaDB · Git · HTMX · REST API design · document automation (python-docx) · command line — enough depth for credible conversations with engineers.

Let's talk

Open to implementation, enablement, internal-tools, and applied-AI roles in the Boston area. In-office or hybrid preferred; open to remote.

Email me Book a call LinkedIn GitHub