Michael E. Marin

Mr. Cooper Group — Dallas, TX

• •Supported agentic AI products for customer service automation across Azure DevOps and MySQL.
• •Participated in product experimentation, user validation, and AI feature evaluation activities.
• •Collaborated with product, engineering, and business stakeholders on AI and machine learning initiatives.
• •Worked in a cross-functional environment translating business requirements into technical deliverables.

Personal Applied-AI / Retrieval Engineering Project

• •Built a production-style RAG platform with FastAPI, PostgreSQL/pgvector, Redis, asynchronous ingestion, hybrid retrieval, reranking, Prometheus metrics, and a React operations console.
• •Offline benchmarked BM25, all-MiniLM-L6-v2 dense retrieval, RRF, and a cross-encoder trained only on SciFact's disjoint training split across 5,183 documents and 300 held-out queries; RRF led with 0.8059 Recall@10 and 0.6816 nDCG@10.
• •Built reproducible serving evaluation and a 30-query mixed load study of the local-provider stack: 300 requests each at concurrency 1, 4, and 8, with 54.73 / 137.22 / 252.24 ms p95 and no failed measured requests; generation was excluded.

Evidence and limitations · Mixed-query load matrix · Serving-path baseline · github.com/xmike04/RAG_OPS · SciFact Benchmark Report · Historical Single-Query Load Summary

University of North Texas · Senior capstone (NASA PACE / UMD collaboration)

• •Led a five-person UNT senior capstone team developing real-time computer vision and interactive visualizations for NASA PACE satellite data.
• •Co-designed the exhibit’s fuzzy-logic gestures and built the ImGui hardware-setup interface for prerequisite checks and camera configuration.
• •Connected and oversaw integration between the gesture-processing system and Unity visualization.
• •Met regularly with NASA PACE collaborators; deployment was handled by other team members.
• •The WAVE installation premiered at the Kennedy Center in Washington, D.C., in March 2025.

Contribution scope and verification limits · pace.oceansciences.org/wave_space_to_ocean.htm · UMD coverage of the Kennedy Center premiere

Personal Project

• •Built and shipped a mobile-first, browser-based life simulation game where players age a character year by year through careers, relationships, finances, and procedurally generated life events.
• •Separated deterministic game systems from authenticated OpenAI narrative generation behind a Supabase Edge Function.
• •Validated bounded requests and model output at the server boundary while keeping provider credentials out of the browser.
• •Verified 795 Vitest test cases across 39 files in the October 27, 2026 CI run, plus desktop/mobile browser smoke with controlled service responses.
• •Public frontend at simlyfe.vercel.app; repaired a demonstrated Nano response-truncation failure, then verified five events through age 10 and same-browser reload.

simlyfe.vercel.app · Evidence and limitations · github.com/xmike04/SIMLYFE

M.S., Artificial Intelligence (Machine Learning concentration)

• Applied Artificial Intelligence — search, heuristics, problem solving
• Advanced Machine Learning — supervised/unsupervised learning, deep learning fundamentals

B.S., Computer Science