Collin Youngerman

A fully automated, end-to-end analytics platform built on GCP, Python, Docker, and Power BI that delivers daily natural gas market intelligence from raw API data to executive-ready dashboards, processing 280,000+ data points across 183 pricing hubs. Prior to this, pricing data was manually retrieved and lacked a centralized reporting infrastructure.

HIGHLIGHTS
• Containerized Python ingestion jobs deployed on Cloud Run pull from the NGI API daily, covering 183 pricing points across all major hubs.
• Raw data staged in Cloud Storage, loaded into BigQuery with watermark-based incremental logic and automatic historical catchup.
• Mart views in BigQuery serve a clean, report-ready layer decoupled from raw ingestion.
• Five-page Power BI dashboard covering spot prices, forward curves, regional comparisons, historical trends, and basis analysis.
• Fully scheduled and automated via Cloud Scheduler. The pipeline completes, and the dashboard refreshes by market open each weekday.
• Modular architecture allows new data sources and dashboards to be stood up quickly on the same infrastructure.

An autonomous AI agent that performs end-to-end financial variance analysis, replacing manual "Big 4" style audit workflows. The system ingests financial data, identifies material anomalies, autonomously investigates root causes by emailing stakeholders, parses their responses, and generates audit-ready Word reports, all without human intervention.

HIGHLIGHTS
• LangGraph state machine orchestrates a 12-node agentic workflow with conditional routing and loops: data ingestion, anomaly detection, stakeholder outreach, response parsing, and report generation.
• Autonomously composes and sends investigation emails via Gmail API (with Outlook and SMTP alternatives), monitors inbox for replies, and extracts root-cause explanations.
• Clarity scoring evaluates stakeholder response quality using LLM-based confidence extraction with heuristic fallback, determining whether explanations are sufficient or require follow-up.
• Generates formatted, audit-ready variance reports as Word documents (.docx) with findings, root causes, and recommended actions using python-docx.
• Full autonomous loop with graceful degradation: every LLM call has a deterministic template fallback, and all external integrations support mock mode for demos.

An agentic procurement system built on Google ADK powered by Gemini. Inventory burns in real time, reorder points trigger automatically, purchase orders are created, high-value POs are routed to a human for approval, and orders are tracked through delivery. It works as a digital twin, a process automation layer, or a testing environment for procurement strategies. Procurement teams spend a lot of time monitoring stock and manually starting orders. This eliminates that while keeping human sign-off where it matters.

HIGHLIGHTS
• Sequential Agent made up of 5 individual agents with specific tasks, wrapped inside a Loop Agent. Execution order is enforced by structure, not by the model's judgment.
• Discrete structured steps shift the system toward deterministic behavior. In a demo, that difference is interesting. In production, it's everything.
• Lean reorder math calculates order quantities dynamically, covering supplier MOQ plus lead-time demand to prevent stockouts.
• Human-in-the-loop approval for high-value orders via tokenized webhook.
• Shares a SQLite database between simulation and agent, bridged by a restock_requests table acting as an event queue.
• Scales from 1 SKU to a full 20-part drone BOM with no core logic changes.