Client Case Studies, Products and Open Source

AI systems I led or built for clients, described without naming the client; the products I built; and public open-source work. Each entry runs from the business problem to what was built, the architecture and AI pattern behind it, and the outcome and my role.11 Client Case Studies · 2 Products · 1 Open-Source Project

Outcomes first. Clients stay anonymous.

Client work is described by its business problem, architecture, stage and outcome, never by the client’s name. ABACare, Tellgence and the open-source projects are named and linked. Every figure on this page is one I can evidence; no product revenue or unverified accuracy claims are shown.

Status:
Industry:
Pattern:

Client case studies

Engagements I led or architected. Clients and their products are not named; the business problem, architecture and outcome are.

A coordinator delegates to specialist agents, a knowledge tool and customer MCP servers
Production platform

Support AI That Resolves Tickets: Ground-Up Rebuild

Software that answers a company’s support tickets, resolves the routine ones itself and hands the rest to people

Multi-agent
RAG
Human-in-the-loop
MCP
Deterministic guardrails
Full case study
Natural-language questions use schema context, access boundaries and SQL validation before returning audited results
Live · 50 Locations

Plain-English Pharmacy Analytics Across 50 Locations

Plain-English answers from pharmacy operational data, without open SQL access

SQL agent
Human-in-the-loop
Deterministic guardrails
Full case study
Every channel joins one customer thread
Production deployment

AI-Native Conversational Commerce Platform

One customer thread across every channel, with AI autonomy set per organization

RAG
Human-in-the-loop
An accepted quote passes idempotency, quote binding, 3D Secure and signed-webhook checks
Deployed

Embedded Payments and AI Risk for Vertical SaaS

Payment acceptance, 3D Secure and AI risk tooling that ERPs and vertical software can embed

Deterministic guardrails
Human-in-the-loop
An intent router sends each email to a specialist chain with order, tracking and store-policy tools
Production Portfolio

E-commerce Support Agent That Answers Order and Refund Emails

Order, tracking, refund and subscription emails answered under each store’s own policy

Multi-agent
Human-in-the-loop
Deterministic guardrails
Full case study
A deep-research agent decides between live web search and a curated knowledge base, streams its sources and stops within a recursion budget
Client delivery

AI Agent Suite for a Life-Sciences Venture Builder

More than 15 specialist agents for founders, led by a deep-research agent that streams its sources

Multi-agent
RAG
An extraction agent drafts and a QA inspector critiques
Delivered · 5.0 client review

AI Document Automation for Construction Accounting

Emailed invoices, insurance certificates and field logs, extracted, checked and routed without hand keying

Multi-agent
Human-in-the-loop
Deterministic guardrails
Expert-vetted templates and the user’s own documents ground each validation draft, which an expert revises and approves under an audit log
Client delivery

AI Validation Documentation for Medical Devices

Template-grounded drafting of validation deliverables, with a qualified reviewer in control

RAG
Human-in-the-loop
Researcher, writer and editor agents cycle on editor feedback under a retry limit, and an evaluation harness scores each article before authors finalize it
Client delivery

Multi-Agent Technical White Paper Generator

Researcher, writer and editor agents in a private workspace, measured by an evaluation harness

Multi-agent
RAG
Human-in-the-loop
API and TLS recovery, mobile checks and release governance converge on TestFlight beta delivery and technical handover
Beta Delivered

AI Health-Coach App Recovery

Interim CTO: production recovery, release governance and a TestFlight beta in three months

React Native / Expo
NestJS / Prisma
PostgreSQL / TimescaleDB
Research runs only when a case needs it
Client Delivery

Corporate Credit Case-Analysis AI Agent

Risk, recovery outlook and draft collection mandates, self-critiqued before expert review

Human-in-the-loop

Products I built

ABACare, my product now going to market, and Tellgence, a Pendoah product I built. Each links to its live site.

Authorizations, session notes, claims and denial follow-up stay connected from first visit to payment
Preparing design-partner pilots

ABACare

Helps ABA therapy clinics get paid for every session they deliver, from authorization to claim

Human-in-the-loop
MCP
Deterministic guardrails

abacare.ai

A caller speaks to a realtime voice agent with turn detection and barge-in
Beta · first business customers live

Tellgence

Voice AI that answers a small business’s phone line, books the appointment and hands off to a person

Voice
Human-in-the-loop
Deterministic guardrails

tellgence.ai

Full case study

Open source

Public code under github.com/shmindmaster.

One canonical set of skills, MCP servers and policy is written into each coding agent’s native configuration
Open Source · Apache-2.0

AgentHub

Keeps every AI coding agent on your machine running the same skills, MCP servers and policy, and tells you when one drifts

MCP

Open-source repositories

Open-source developer tooling I maintain under github.com/shmindmaster.