Kevin Badinger
Principal Software Engineer — AI Systems · Shipping production LLM systems
Greater Philadelphia · Remote (US) · kevin@kevinbadinger.com · kevinbadinger.com · linkedin.com/in/kbadinger
Summary
Principal software engineer and hands-on architect who ships production AI systems end to end — autonomous AI agents, multi-agent workflows, retrieval-augmented generation (RAG), multi-LLM orchestration, evaluation harnesses, and human-in-the-loop systems — in Python and TypeScript on AWS. Founder and operator seasoning underneath the code: built and ran a nurse aide credentialing platform whose testing protocols became the standard in multiple states, as its original architect — 1M+ certifications across 35+ states, 99.9% uptime, zero known breaches — leading and developing 80+ engineers over the tenure — ~30 at any one time across U.S. and offshore teams — while staying in the codebase. Also ran the whole business as Chairman, President and Managing Partner of Credentia — payroll, IT and operations for a ~600-person organization at peak, mostly field nurses. Builds reliable AI systems daily with the same production rigor: real constraints, tests, observability, and rollback. Author of The AI Reckoning (forthcoming), on production AI in regulated industries.
Experience
Founder, President and Principal AI Engineer
Resolvent Technologies · Remote (Pennsylvania) · resolventtech.com
Hands-on production-AI engineering practice; builds and operates AI systems and platforms across regulated and enterprise domains.
Advisory and cleanup engagements across healthcare, finance, logistics, transportation, and state government — evaluating AI-built and legacy systems and rebuilding what shipped before anyone understood it.
Retained as a technical expert witness in technology litigation — expert reports and depositions.
Production AI and Engineering Practice (2023 – Present)
- Ship production AI systems end to end — autonomous AI agents, multi-agent workflows, retrieval-augmented generation (RAG) pipelines, MCP servers, and multi-LLM orchestration across Anthropic Claude, OpenAI, Gemini, and Grok — in Python and TypeScript on AWS, including Guardian (a persistent-memory agent platform with a knowledge graph), ThisWeekEats (~130k LOC), ReadReceipt, and PenForge.
- Built and direct Ralph, an autonomous AI developer agent, building Qtown (qtown.ai), a polyglot distributed-systems simulation spanning 12 languages and 9 microservices (Go, Rust, Python, LangGraph, TypeScript, Kafka, gRPC, GraphQL, ~100K lines of code), through story-level acceptance criteria, per-task model routing, code review gates, and an audit and verification loop; the agent wrote 88% of v1 across 1,451 commits and still writes most of v2.
- Built a multi-LLM adversarial evaluation panel with human-in-the-loop gating — four independent model votes with research grounding, an adversarial refutation pass, confidence-weighted aggregation, and human sign-off before any decision ships.
- Shipped Charted (charted.work), a free public career-discovery engine built on O*NET data — production cutover on Vercel and Railway with a custom domain, no account required.
- Published HIPAA Safe Harbor de-identification tooling (hipaa-de-id, public repository) — detects and transforms all 18 Safe Harbor identifier categories across text, CSV, JSON, FHIR R4, and HL7v2, with a PHI-stripping proxy for LLM API traffic and an encrypted re-identification store with audit trail.
- Built an LLM-judged voice-fidelity evaluation (letter grade, 0–100 score, itemized fixes) with a learn-from-edit loop that extracts semantic style rules from user edits into a persisted per-user model.
- Built a public evaluation and monitoring gate for PenForge's LLM grader — a golden test set with known answers, model-judged scoring, a CI check that blocks releases when accuracy drops, and per-call cost and latency tracing with OpenTelemetry.
- Deployed Charted's scoring service on Amazon Web Services — Bedrock, Lambda and ECS, provisioned with Terraform.
- Run ThisWeekEats on a managed Kubernetes cluster (DigitalOcean) — containerized services, rolling deploys and autoscaling.
- Operate these systems with production discipline — data pipelines, evaluation harnesses, observability, and rollback — applying modern SDLC and AI-assisted development to ship quickly without sacrificing quality.
Rail Track-and-Trace Platform and RailVantage (Architect and Builder, 2012 – Present)
- Operate the live rail track-and-trace platform I built for a rail-logistics client, running real operational data from Class I railroads — real-time integrations with two Class I railroads and a waybill feed from a third, sub-3-second routing across 10,000+ point networks, and offline-first Flutter field apps for rail crews.
- Extended it in 2026 into RailVantage, a mobile-first rail-operations product that absorbed a railroad car repair billing system (an AAR-mandated market), now live with real waybill data.
Founder, Chief Technical Architect, Chairman, President & Managing Partner
Credentia / Conductor Platform (engineering-led B2B SaaS)
Architected and built the Conductor platform from 2002, bringing it into nurse aide credentialing in 2004; formed Credentia in 2014 as the operating company and served as Chairman, President and Managing Partner through 2023 while remaining Conductor's hands-on chief technical architect throughout. Bootstrapped, profitable from year one, no outside funding; exited 2023, with platform support winding down through 2024.
- One continuous platform grown from point solutions into the system powering a nationwide U.S. nurse aide testing and credentialing program: 1M+ certifications across 35+ states, 99.9% uptime, zero known breaches.
- Built and launched CNA365, a nurse aide credentialing system that moved every state program off the incumbent's legacy systems in about six weeks (January–February 2022), with online applications, electronic training verification, and an online remote-proctored written exam at no fee increase.
- Became the official test administrator for the national nurse-aide exam in 2022 after buying Pearson VUE's nurse aide testing business in 2021; administered 65,000+ skills exams and 54,000+ written exams in 2022 across 13 states and U.S. territories, including state registries.
- Led a multi-team engineering organization across the U.S., India, Serbia, and Brazil — ~30 engineers at any one time, with 80+ hired and developed over the tenure — running day-to-day engineering directly across all time zones; set code-review, design-review, and engineering standards; sub-5% annual voluntary engineering turnover across the run.
- Operated real-time data pipelines and a published external API layer (authentication, scheduling, scoring, reporting) on Microsoft Azure, integrating candidate-management systems, credential-verification services, and state registries: bidirectional exchange on a regulated, high-availability platform with strict data consistency and auditability.
- Engineered for deliberate demand spikes: quarterly exam-offer releases at 7 PM in each nurse's local timezone created thundering-herd load, solved in stages — load balancing, client-side refresh caps, per-user throttling across devices and sessions; at peak, 500 nurses filled thousands of offers in under 10 seconds.
- Ran human-in-the-loop operations at scale: a proctor and evaluator workforce executing high-stakes certification exams, recurring audit programs (humans auditing the human scorers), and automated escalation with human fallback.
- Met the compliance bar of state contracts and their RFPs: SOC 2 and PCI-DSS attestations, multi-state regulatory compliance, and a HIPAA-adjacent regulated environment handling state-designated PHI.
- Shipped SCORE, a mobile OCR scoring application that turned paper answer sheets into structured, machine-scored data in the field, feeding the automated credentialing pipeline.
Lead Engineer
StarFloors — Flooring-Installation and Senior-Living Maintenance Platforms (Resolvent client)
- Owned end-to-end delivery of two enterprise SaaS platforms over nine years in production — a general flooring-installation platform and a senior-living maintenance platform — serving major senior-living operators, plus a Fannie Mae property-services line.
- Built the senior-living maintenance platform around automated room turns — community floor plans in-system; selecting a room triggered bid requests, contractor scheduling, and next-day completion — covering construction-grade maintenance (flooring, painting, construction repairs) across communities.
- Designed the automated workflow engine (request to approval to installation to payment to survey) with Salesforce/CRM and bidirectional QuickBooks integrations; 85% worker-productivity gain, 53% faster turnaround, 10K+ daily transactions across the business.
Other Client Platforms
Resolvent Technologies
- Built and support IntelliSite, the document-management platform for a national RV-resort operator — a client engagement ongoing since 2005.
- Built systems for small and mid-size businesses — a global shipping integration for a reshipper and plant-nursery management — and ran SEO that moved 50+ sites to Google's first page for 20+ clients.
Principal Software Architect
Trintech (Enterprise Financial Software)
- Principal Software Architect, ReconNET LCM — banking reconciliation and compliance platform serving 300+ Fortune 1000 financial institutions, with clients including a top US bank, a regional bank, a major US investment firm and national retailers; primary architect of LCM Payments (.NET) with real-time fraud prevention and behavioral anomaly detection on live payment streams.
- Designed a custom object-relational mapping layer supporting SQL Server, Oracle, Informix, and DB2 behind one abstraction; 100x performance improvement via SQL Server and Oracle tuning (published work); reversed sales win/loss from 0% to 100% in one quarter through a GUI redesign.
Earlier
IBM, Sales Engineer; Texas Instruments, Software Engineer; DSC Communications, Senior Software Design Engineer
- DSC Communications: built embedded broadband cross-connect test systems that cut manufacturing test costs 50% (~$1M/year savings, ~$500K per installation).
- Texas Instruments: converted an IMS mainframe semiconductor machine-tracking system to a distributed object-oriented architecture (Smalltalk, GemStone), tuned to 40 transactions/second; 24x7 support across 12 customer sites.
Major Platforms Built
Conductor
Nurse aide credentialing platform. In production for two decades. 1M+ certifications across 35+ states. Testing protocols adopted as the standard by multiple states; clients operated on it daily. Multi-portal architecture. 50+ state database integrations. Bootstrapped to a successful exit in 2023.
ReconNET LCM (Trintech)
Banking compliance platform for 300+ Fortune 1000 financial institutions. Bank reconciliation. Cash forecasting. Unclaimed property management. Payment lifecycle. Multi-national deployment.
Personality-Assessment Platform
Assessment platform for a personality-assessment company. In production for 22 years (1998–2020). Seven-part communication-style assessment. Multi-national deployment. Complex grading algorithms.
SCORE
Mobile exam platform. iOS/Android native apps. $200k+ annual savings. Offline exam administration. Instant score delivery.
StarFloors
Senior-Living Maintenance Platform — construction-grade maintenance with automated room turns. 85% productivity gain, 53% reduction in work-order processing time. 10K+ daily / 5,000+ monthly transactions. Deployed to major U.S. senior-living operators. 9 years operational.
AI Projects
All projects built using Claude Code as primary development environment.
AI Agent Audit
Production audit daemon for OpenClaw AI agent installations. 7 always-on monitors, 20 periodic audit sweeps, auto-remediation, HTML reporting. Configuration assessment, credential exposure scanning, marketplace/skill supply chain checks.
HIPAA De-Identification Pipeline
Open-source HIPAA Safe Harbor de-identification system for healthcare AI applications. Automated detection and removal of all 18 PHI identifiers from clinical text per 45 CFR 164.514(b). Enables compliant AI model training and inference on protected health data. Built because most healthcare AI startups skip de-identification until it's a breach.
Guardian
Always-on AI memory and decision system. MCP server exposing 100+ tools, knowledge graph, RAG over hybrid search (sqlite-vec + FTS5). Orchestrates Claude, OpenAI, Gemini, and Grok. Dreamer consolidation engine derives facts with lineage; human-in-the-loop review gates what reaches model context.
ThisWeekEats
AI meal planning system with Stripe-integrated subscription payment processing. Constrained generation with hard rules for dietary preferences, time limits, favorites tracking. 50 real beta testers. Planning time reduced from 2-3 hours to under 10 minutes.
DebatePanel (formerly DecisionForge)
Multi-LLM debate engine for complex decisions. Evaluation harness with rubric-based scoring. GitHub Actions CI/CD.
Skills
Core Competencies
Technical Leadership and Mentorship, Production AI Systems, Software Architecture and System Design, Evaluation and LLM-Judge Scoring Harnesses, Observability, Retrieval-Augmented Generation (RAG), Autonomous AI Agents, Multi-Agent Workflows, Multi-LLM Orchestration, Distributed Systems, AI-Assisted Development, Regulated Healthcare Data (PHI De-identification, Audit Trails), Human-in-the-Loop Systems, Data Pipelines and Orchestration, API-Driven Architecture, Code Reviews and Design Reviews, Engineering Standards and Best Practices, Cross-Functional Technical Leadership, Build-vs-Buy and Product Strategy, Regulated SaaS, High-Availability Systems
Technical Skills
Python, TypeScript, JavaScript, AI Agents and Tool-Calling, LLM Evaluation and Scoring Harnesses, Retrieval-Augmented Generation (RAG) Pipelines, LangChain, LangGraph, MCP Servers, Amazon Web Services (AWS), Continuous Integration and Continuous Delivery (CI/CD), Anthropic Claude, OpenAI, Gemini, Grok, C# and .NET, Vector Databases (pgvector, Qdrant, Pinecone, ChromaDB), Microsoft Azure, React, Next.js, Node.js, SQL, PostgreSQL, SQL Server, Docker, Kubernetes, Terraform, Go, REST APIs
Key Metrics
Education
Bachelor of Business Administration (BBA), Information Systems — Baylor University
Certifications
- HIPAA-trained (certificate on file) — privacy, security, and breach notification rules. Built and operated production systems in HIPAA-adjacent regulated environments; published HIPAA Safe Harbor de-identification tooling.
Current Work
Building AI-powered systems for healthcare and regulated industries. Primary development environment is Claude Code — all current projects built using AI-accelerated development workflows. Focused on production deployment of LLM capabilities with HIPAA-adjacent compliance, data security, and governance built in from day one.