full two-page version →

Kelly Michels

Houston, TX (Remote) · kelly@evomedia.net · 713-412-9323 · Project lists and case studies: evomedia.net

Multi-Tenant SaaS & AI / RAG Engineering · Full-Stack Developer (Python / Node.js) · Principal QA Engineer / SDET · AWS Cloud Infrastructure

Summary

Builds and operates production multi-tenant SaaS and AI systems end to end — schema and API design through containerized AWS deployment and monitoring. Currently running a multi-tenant Retrieval-Augmented Generation platform in production, the application platform beneath it, and products shipped on top of both. Takes an AI feature from “we should try this” to something customers depend on: retrieval that cites its sources, guardrails with regression suites behind them, per-tenant isolation that holds. Backed by enterprise quality engineering at Autodesk on Maya and 3ds Max, including AI-driven animation features taken from first build to global release.

Current Work (built and operated solo — sign in to any app as demo@evomedia.net; project names shown are internal working names, not final)

Multi-tenant RAG platform — in production. Answers only from a tenant's own records, with citations, and declines the rest. Model-agnostic per tenant, fully local via Ollama for regulated customers. Hybrid retrieval over Qdrant plus text-to-SQL verified against ground truth. Four layered guardrails — grounding policy, a pre-LLM relevance gate that declines off-topic and adversarial questions before the model is called, a SELECT-only tenant-scoped SQL boundary, and propose-never-execute actions. Live as “Ask AI” inside the EHS and civil-code products. FastAPI · LlamaIndex · Qdrant · LiteLLM / Ollama

evo.platform — the foundation beneath the products: authentication, row-level tenant isolation, and an API designed as the core rather than bolted on. A shared SDK ships a new product in days, not months. Being prepared for open-source release. NestJS · Prisma · PostgreSQL · JWT + WebAuthn · Stripe

Multi-tenant compliance SaaS for EHS industry — multi-plant manufacturers. Onboards a facility from a company name alone, auto-resolving its EPA ID and importing permits from EPA and Texas TCEQ. Permit lifecycle, OSHA incidents, passkey sign-in, Stripe billing. Python · NiceGUI · PostgreSQL · SQLAlchemy · Docker · AWS EC2

evo.provensheet — turns the spreadsheet a business runs on into a shared web app and proves the numbers survived: every formula rechecked against Excel's own cached result. Built-in sample: an 18-sheet estimating workbook, 5,121 cells, all matching. Next.js · IronCalc (WASM) · Dexie · Prisma · PostgreSQL

QA for AI systems, and open source — a documented defect loop against text-to-SQL, retrieval and guardrail failures, every defect with a reproduction, regression test and ground-truth verification, plus a guardrail regression suite and LLM-as-judge cases, one per closed defect. evo.zscripts and evo.locate are open source. It all runs on one nginx edge with self-hosted mail, Git and metrics — a single AWS box, no managed PaaS. pytest · PowerShell · FastAPI · Docker · Prometheus / Grafana

Professional Experience

Principal Software QA Engineer / Senior QA Engineer · Autodesk

2006 – 2026

Core QA contributor for Autodesk Maya MotionMaker AI and the ML Toolkit, from first build through production release of AI-driven animation workflows.
Contributed to MotionMaker AI earning the Customer Experience Award at TechX 2026; supported executive demo readiness for TechX 2025.
Led QA strategy and execution for flagship Autodesk Maya and 3ds Max — regression, integration, E2E and visual validation.
Built automation and API testing in Python, PowerShell and AI-assisted workflows; integrated into Jenkins CI/CD with Artifactory. Many years of Agile delivery across two-week sprint cycles.

Core Skills

AI & RAG Retrieval-Augmented Generation, hybrid retrieval, embeddings, text-to-SQL, citation enforcement; LLM guardrails — relevance gating, prompt-injection defence, read-only SQL boundaries, guardrail regression suites, LLM-as-judge evaluation; LlamaIndex, Qdrant, LiteLLM, Ollama · Multi-Tenant SaaS row-level isolation, shared-platform & SDK design, API-first architecture, JWT + WebAuthn, Stripe · Full-Stack Python / FastAPI / NiceGUI, Node.js / Next.js / NestJS, TypeScript, React, PostgreSQL, Prisma; async job queues and schedulers (RAG platform: in-process asyncio worker, broker seam for Celery / ARQ) · AWS & DevOps EC2, CloudFront, Route 53, Docker, nginx, Jenkins, Prometheus / Grafana