Recent work.
Selected AI systems, data infrastructure, and applied research from recent client engagements across education, philanthropy, government, and mission-driven organizations.
AI systems and applications
embedded ai application development for public school districts
K-12 Public Education
Ongoing engagement where we embed with instructional and operational leader to identify the highest-leverage teacher and staff workflows. Each district’s build is unique to its instructional priorities and operational context.
Stack: Custom AI application platform, Claude and Gemini APIs, workflow - specific agent design
governance policy chatbot
Mission-driven consulting firm
A Claude API RAG chatbot for operationalizing organizational AI governance policy. Shifted staff engagement with governance documentation from near-zero direct reference to active, ongoing use.
Stack: Claude API, RAG, Custom retrieval
AI Adoption hub
mission-driven consulting firm
An internal hub where staff submit AI tools, skills, and use cases for organizational adoption, plus request new builds by describing the problem and their envisioned solution. Submissions and finished builds move through a tiered review pipeline: staff testing and voting first, executive approval second.
Stack: Claude API, custom review pipeline, web application
custom case management system
family foundation
A custom case management system built on Google Workspace as a zero-cost, org-owned alternative to commercial case management platforms. Spans 16 record types and 8 reporting rules across the organization’s full program portfolio. Expanding with a natural - language query layer and automated grant-report narrative generation.
Stack: Google Workspace, Google Apps Script, Claude API for LLM query layer
custom mcp server for financial screening
Financial Services
A Model Context Protocol server that wraps a public data source for automated screening workflows. Six-tool architecture with attributed-fact data structures, fallback logic for inconsistent source at a, and a three class error taxonomy that traces every numeric output to its origin. Paired with a Claude skill that produces first-pass analytical memos.
Stack: Python, Model Context Protocol, Claude Agent SDK, Claude Skills
ai adoption hub
mission-driven consulting firm
An internal hub where staff submit AI tools, skills, and use cases for organizational adoption, plus request new builds by describing the problem and their envisioned solution. Submissions and finished builds move through a tiered review pipeline: staff testing and voting first, executive approval second.
Stack: Claude API, custom review pipeline, web application
staff capacity planning app
Mission-driven consulting firm
A custom Google Workspace application for staff capacity planning. Integrates with staff Google Calendars to capture meetings and produces capacity reports through a conversational workflow with energy weighted task categorization.
Stack: Google Workspace, Google Apps Script, Google Calendar API, conversational AI layer
deep dives: research and program evaluation.
Government mental health response program
A government funded 911 alternative mental health response initiative
THE SITUATION: The initiative launched to provide an alternative to police response for mental health crises. Leadership needed real-time implementation analysis to understand what was working, what needed adjustment, and how to communicate results to public and government audiences. Data lived across multiple partners (city agencies, clinical response teams, community organizations) with no unified analytical view.
WHAT WE DID: Coordinated data collection across city agencies, clinical partners and community stakeholders to build a shared analytical foundation. Ran ongoing implementation analysis surfacing what was working across dispatch, response, and follow up. Co-designed performance dashboards translating complex administrative and equity data into accessible tools for public and government audiences. Worked directly with program leadership to translate findings into strategic decisions.
WHAT SHIPPED: Real time implementation analysis system, performance dashboard for internal and public audiences, analytical reporting, and strategic recommendations for program leadership.
WHAT CHANGED: Program leadership gained a shared view of implementation that let them coordinate across partners and make decisions with real data. Public dashboards gave community members and government audiences direct access to how the initiative was performing without needing to interpret raw administrative data.
workforce development data system
a workforce development organization serving multiple funder-defined program cohorts
THE SITUATION: The organization was managing multiple funding streams with different reporting requirements. There was no unified view of program performance. Quarterly reporting errors averaged seven or more per cycle. Staff time was consumed reconciling data across systems and preparing reports rather than serving participants.
WHAT WE DID: Designed and implemented an organization-wide data system serving 45+ staff members plus board and funder reporting functions. Built for monthly visibility into program performance and compliance across multiple funding streams. Established data governance standards defining who owned the data, how it flowed between systems, and what quality checks ran at each stage.
WHAT SHIPPED: Production data system with staff facing dashboards, board reporting infrastructure, funder-specific reports aligned to grant requirements, dat governance documentation, and staff training materials.
WHAT CHANGED: Quarterly reporting errors dropped from seven or more per cycle to zero. Staff time shifted from data reconciliation to program work. Board and funders gained monthly visibility into program performance that previously required manual assembly. The organization’s compliance posture across multiple funding streams became defensible without individual heroics.
housing and supportive services impact analysis
a housing and supportive services organization advocating for policy changes based on program outcomes.
THE SITUATION: The organization operated housing and supportive services programs. The team knew the work was changing lives. What they lacked was quantitative evidence to bring to policymakers. Existing administrative data was rich but disconnected in ways that made it hard to tell the story.
WHAT WE DID: Directed data strategy and reporting for housing and supportive services programs. Built dashboards and impact analysis designed specifically for funder and policy audiences. Conducted Python based analysis linking criminal history records to housing outcomes to surface program effects that weren't visible in housing status alone. Translated findings into policy briefs and testimony-ready materials.
WHAT SHIPPED: Analytical infrastructure connecting administrative datasets, impact dashboard, policy briefs, and evidence packaged sized for legislative testimony.
WHAT CHANGED: Findings from analysis were used directly in policy advocacy. The organization gained a defensible evidence base for the impact claims they were already making, connecting the direct experience of participants to qualitative outcomes that changed how funders and policymakers understood the work.