Enterprise AI
The engineering behind the local work
Six years of AI engineering experience across retrieval, agent orchestration, model deployment, and optimization. That background informs how we scope and test small-business systems. Larger technical engagements are available by agreement; projections below are labeled separately from measured outcomes.
How engagements work at this end
Capabilities
What gets built
Agentic system design & build
Multi-agent pipelines: intent resolution, tool orchestration, external integrations, and memory that survives across sessions.
RAG & knowledge infrastructure
Hybrid retrieval over your documents — vector search for meaning, graph search for relationships — at corpus sizes in the thousands and up.
Computer vision & geospatial
Detection, tracking, and the post-processing that turns raw model output into something a person can act on.
Data & training pipelines
Dataset construction, labeling strategy, and the quality controls that keep a dataset from quietly capping your model ceiling.
MLOps & model lifecycle
Experiment tracking, hyperparameter search, registry, and automated retraining on SageMaker, MLflow, and Kubeflow.
Optimization & decision systems
Constraint satisfaction and classical operations research for scheduling, assignment, routing, and batching.
- PyTorch
- XGBoost
- scikit-learn
- pandas
- GeoPandas
- OR-Tools
- LangGraph
- Google ADK
- MCP
- Neo4j
- DSPy
- AWS SageMaker
- AWS Bedrock
- MLflow
- Kubeflow
- Docker
- Kubernetes
- React
Case studies
Selected systems and engineering work
Hybrid RAG
Contract Intelligence at Scale
An ingestion and retrieval pipeline over thousands of contract documents, combining vector and graph search into one hybrid system so both semantic questions and relationship questions could be answered from the same corpus. Built on AWS Bedrock and DSPy with AuraDB for the graph layer.
1,000s
Documents ingested
Vector + Graph
Retrieval modes
Agentic product
No-Code UI Generation Agent
A RAG-backed agent workflow that walks a user through requirements discovery, then generates a polished prototype grounded in user research, style guides, and images. Users never touch code. Led the team of four building the agentic backend and the web application it runs in.
2 mo → days
Inception to prototype
90%+
Timeline reduction
Multi-agent orchestration
Technical Agent Orchestration
An agentic system that accepts arbitrary prompts and data, infers intent, and routes to the right capability — object detection, track generation, or external tool calls for enrichment. Built from scratch on Google ADK with a custom traceability module persisting native memory to Neo4j.
80%+
Projected operator time savings
Neo4j
Cross-session memory
Healthcare logistics · hybrid ML + optimization
Transportation Assignment Optimization
An optimization architecture assigning transportation rides across available capacity — a hybrid of XGBoost regression and OR-Tools constraint satisfaction, with models managed on SageMaker and MLflow and retrained automatically through CI/CD.
$70M+
Annual savings potential
XGBoost + CSP
Hybrid approach
Computer vision · geospatial
Wildfire Fuel Segmentation & Fire Prediction
Led a team deploying three computer-vision models for wildfire fuel segmentation and fire behavior prediction. Architected a ground-truth validation framework over satellite imagery, modified segmentation architectures in PyTorch to beat baselines by 15%, and automated the pipeline with Kubeflow — cutting five weeks from the training cycle.
3 models
Deployed to production
+15%
Over baseline
Multi-agent · production infrastructure
Multi-Agent GenAI Command & Control
A scalable multi-agent, multi-system GenAI orchestration platform deployed as microservices on Docker and Kubernetes. Hybrid retrieval spanning GraphRAG, vector search, and agentic text-to-SQL; secure MCP-based tool calling; and a dual-layer memory system.
3-mode
Hybrid retrieval
MCP
Secure tool calling
Depth
Where the specialization actually is
Agentic AI
Multi-agent pipelines designed from the ground up rather than assembled from a template — intent resolution over arbitrary prompts and data, dynamic tool selection, and orchestration across external services, with custom traceability that persists agent memory to a graph database for real cross-session continuity.
Retrieval & knowledge systems
Hybrid retrieval combining dense vector search with graph traversal, so a question can be answered by both semantic similarity and the explicit relationships between entities. Proven on document sets in the thousands.
Computer vision
End-to-end vision model development: dataset construction, training, evaluation, and the post-processing stage where detections become tracks, entities, and decisions.
Model lifecycle & MLOps
Managing models across the full training lifecycle — hyperparameter tuning, experiment tracking, registry, deployment — on AWS SageMaker and MLflow, with Kubeflow pipelines for automation and CI/CD for monitoring and retraining.
Optimization & classical AI
Constraint satisfaction and operations research for problems where the optimal answer is computable rather than learned — assignment, batching, scheduling, and routing.
Demo
A pipeline you can watch run
Computer vision · side project
Real-Time Fighter Action Recognition
A two-stage vision pipeline on Super Smash Bros. Ultimate gameplay: stage one detects and crops the fighter from the raw frame, stage two runs 36-way action classification on the crop. The clip shows recorded model outputs on real gameplay video.
Background
Experience & credentials
Staff AI Engineer
Fortune 100 aerospace & defense company · Feb 2026 — Present
Team lead across generative AI efforts: a UI-generation system grounded in user research and style guides that cut inception-to-prototype time from two months to days, and an agentic data-workflow platform projected to save operators over 80% of their time.
Senior AI Engineer
National healthcare transportation company · Aug 2025 — Feb 2026
Built an AI-driven transportation optimization architecture with $70M in potential annual savings (XGBoost + OR-Tools constraint satisfaction), managed models on AWS SageMaker and MLflow with automated retraining via GitLab CI/CD, and developed an agentic RAG system for contract analysis on AWS Bedrock, DSPy, and AuraDB.
Senior AI Engineer
Fortune 100 aerospace & defense company · Jun 2023 — Aug 2025
Co-led a scalable multi-agent GenAI command-and-control system — hybrid GraphRAG + vector + agentic retrieval, MCP tool-calling, dual-layer agent memory — and led a team of engineers deploying three computer-vision models for wildfire fuel segmentation and fire behavior prediction.
Leadership Development Program
Fortune 100 aerospace & defense company · Jan 2020 — Jun 2023
Selective rotational program for high-potential technical leaders: explainable-AI tooling (custom PyTorch Grad-CAM), dataset-bias dashboards, and software/systems engineering across four defense programs.
Johns Hopkins University
M.S. in Artificial Intelligence. In progress — estimated completion fall 2027.
University of Colorado, Colorado Springs
B.S. in Mechanical Engineering — Special Honors, 3.9 GPA. Minors in Electrical Engineering and Aerospace Engineering. Colorado school, Colorado degree.
Have an AI problem at a larger scale?
Architecture reviews, prototypes against your real data, and embedded delivery. Describe the system and the constraints.