Portfolio — 2026
Christopher
Millones.
AI/ML Engineer
I build AI systems that actually ship — LLM pipelines, RAG architectures, and the MLOps wiring around them, running at production scale in regulated environments. Four years of making models reliable enough to trust with real decisions.
01 — About
Most AI projects fail in the gap between the prototype and production. I work in that gap.
Four years of it, specifically — in healthcare, where the margin for error is low and the data is messy. I work across the full stack of an LLM system: prompt design, retrieval architecture, evaluation pipelines, and the cloud infrastructure underneath.
The goal is always the same. Models that perform consistently when it counts, and a paper trail that explains why when they don't.
- 4+Years Experience
- ~2MAnnual LLM Invocations
- 84%Cache Hit Rate Achieved
- 70% → 90%Extraction Accuracy
02 — Skills
The tools are the easy part. Here's what I reach for.
AI / ML
01Cloud
02Languages
03Frameworks
04MLOps / DevOps
05Compliance
0603 — Experience
The roles, the rooms, and the systems that came out of them.
01 — Role
Software Engineer – MLOps & AI
Glidewell Dental Inc.
Nov 2021 – Present
Dental prescription forms are handwritten, inconsistent, and structurally chaotic — and we process about a million of them a year. I designed the pipeline that reads: a hierarchical LLM fallback architecture that handles edge cases gracefully, paired with 8+ rounds of prompt iteration that got us to a 90% extraction accuracy and 84% cache hit rate, and meaningfully reduced per-call costs at scale.
Call center agents were losing time hunting through internal docs for answers that should've been instant. I designed a RAG system that changed that — not a basic vector search, but a full retrieval stack with semantic chunking, query rewriting, and LLM-based reranking to make sure the right answer actually surfaces.
Doctors leave free-text comments on orders. Those comments contain routing decisions that the operations team needs to act on — fast, at scale. I built the pipeline that reads them structurally, extracts the intent, and routes accordingly. About a million orders a year flow through it.
Prompt engineering without measurement is just guessing. I built the system that made iteration data-driven: automated eval against ground truth across relational and warehouse databases, with A/B benchmarking baked in. Accuracy went from 70.2% to roughly 90% over successive rounds.
Beyond the headline projects: CV and OCR model pipelines, production monitoring with multi-channel alerting, a facial recognition access control system, and a full cost-and-performance analysis for migrating our LLM provider from Bedrock to Vertex AI.
02 — Role
Market Solutions Engineer
Star Micronics Inc.
Aug 2019 – Jan 2021
Before AI, I was the integration person — the one vendors called when something wasn't working, the one writing the SDK docs developers actually used. Hardware, retail systems, anything that had to plug into anything else. It's where I learned to talk technically across the stack, and it still shows up in how I work.
04 — Education
B.S. Computer Science · Cal State Fullerton · GPA 3.70
California State University, Fullerton
B.S. Computer Science
GPA
3.70
Graduated
Aug 2021
05 — Contact
Let's build something that ships.
If you're shipping something hard with LLMs, agents, or the infrastructure underneath — I'd like to hear about it. Based in Irvine, CA. Remote or hybrid, both fine.