Portfolio

Case studies in production AI.

Each of these shipped, ran against real traffic, and moved a number that mattered. Problem → approach → impact.

Agentic AI · Walmart Global Tech

Distributed Agentic Monitoring System

SRE teams discovered DC infrastructure incidents reactively — burning hours in the discover → escalate → troubleshoot chain before a human even understood what broke.

Deployed autonomous agents across server infrastructure that continuously track memory, CPU, and storage, detect anomalies, log diagnostic notes, and page SRE teams with pre-analyzed findings.
Idle discovery time eliminated. SREs now arrive at anomaly onset with full context — collapsing MTTR from ~4 hours to under 30 minutes.

PythonMulti-agentAnomaly DetectionSREObservability

Read the deep dive

RAG · Walmart Global Tech

RAG Diagnostics Pipeline

Real-time context for every agent decision

Monitoring agents could see telemetry but lacked the operational and historical context to diagnose why something was failing.

Engineered retrieval-augmented pipelines integrating ServiceNow and internal APIs as LLM knowledge sources, surfacing real-time operational context and prior incident signal at inference time.
Agents reason over live operational state and institutional memory — producing diagnoses grounded in what's actually happening, not just raw metrics.

RAGLLMsServiceNowFastAPIVector Retrieval

Read the deep dive

Machine Learning · Stealth Startup

PTSD Symptom Analysis Model

End-to-end raw data → trained model

An AI-led PTSD symptom analysis product needed a model that could recognize meaningful language patterns from messy, real-world input.

Trained a language pattern-recognition model and built the full pipeline around it — data collection, cleaning, and training-set preparation from raw sources.
Became the core of the analysis pipeline; architecture and decisions were documented for reproducibility and onboarding.

NLPData PipelinesModel TrainingPython

Read the deep dive

Platform · Walmart Global Tech

Messaging Platform Re-architecture

20%+ performance gain · every NA DC

A legacy messaging platform resolved recipients sequentially, creating a hard latency ceiling as the audience grew.

Re-architected recipient resolution as concurrent FastAPI calls, removing the serial bottleneck without changing the platform's external contract.
20%+ performance improvement, adopted across all North American distribution centers and supporting teams.

FastAPIConcurrencyDistributed SystemsPerformance

Machine Learning · simuwatt

Energy-Auditing Model & Data Pipeline

+30% data training set · +20% retention

A PyTorch energy-auditing model — built to benchmark buildings and recommend cost reductions — was starved for quality training data.

Acquired and cleaned building-energy datasets via web-scraping pipelines and Python parsing; automated QA and user workflows with Selenium to free engineering time for modeling.
Expanded the training database 30%, doubled viable in-house data, and contributed to model fitting that raised year-end retention 20%.

PyTorchWeb ScrapingSeleniumData Engineering

Want the deep dive? Get in touch View résumé