AI & Automation

I design AI-assisted workflows that combine automation with structured validation, grounding controls, human review, and operational governance. This portfolio includes hands-on applications as well as AI-enabled process improvements implemented across technical operations and documentation environments.

Featured AI Application

Python · OpenAI API · Streamlit

AI Documentation & Operations Assistant

The Problem

Operational teams routinely work from fragmented incident notes, technical observations, meeting notes, and status updates. Turning that information into consistent documentation requires time while also creating a risk that assumptions or unsupported conclusions can enter the final record.

The Solution

I designed and built an AI-assisted application that transforms unstructured operational notes into structured, reviewable documentation while preserving uncertainty and requiring human validation before the output is treated as complete.

Source Notes → Input Validation → AI Generation → Output Validation → Human Review

Supported Outputs

  • Incident summaries
  • Executive updates
  • Preliminary root cause analysis drafts
  • Standard operating procedure drafts

Architecture

  • Python application architecture with separated prompt, analysis, and validation components.
  • Streamlit interface for document selection, operational input, generation, validation, and review.
  • OpenAI API integration for controlled document generation.
  • Environment-based secret management so API credentials remain outside source control.
  • GitHub-based source control with feature branches and pull-request validation.
  • Cloud deployment connected to the production branch.
Python Streamlit OpenAI API Prompt Engineering Pytest GitHub Actions CI Secrets Management Human-in-the-Loop
Live Working AI application deployed from a tested GitHub production branch.

Responsible AI & Governance

AI Grounding

Designing Against Unsupported AI Conclusions

The Failure I Found

During testing, the model received incident notes stating that application instances were restarted and that error rates later returned to normal. The first implementation described the restart as reducing the error rate even though the source notes never established that causal relationship.

Why It Mattered

In operational documentation, chronology is not proof of causation. An unsupported relationship introduced into an incident summary can later be mistaken for an established technical finding.

The Control

  • Added system-level instructions prohibiting unsupported causal relationships.
  • Required confirmed facts to remain separate from assumptions.
  • Required missing information to be identified instead of inferred.
  • Prohibited invented actions, investigations, decisions, monitoring activities, and outcomes.
  • Required suspected causes to remain explicitly unconfirmed unless supported by the source.

Regression Testing

I reran the same source notes after each control change. A second test exposed another unsupported statement describing monitoring activity that had never been provided. I tightened the grounding rules again and repeated the test until the output preserved the facts without inventing causation or operational activity.

Design principle: AI-generated operational documentation remains a draft. The application surfaces uncertainty and requires explicit human review rather than treating model output as an authoritative record.

Quality Engineering

Testing · CI

Building Automated Validation Into the Delivery Workflow

Testing Strategy

I added automated unit tests around the application’s input and output validation behavior so basic quality controls are verified independently of the AI model.

Automated Tests

  • Valid operational input is accepted.
  • Empty source input is rejected.
  • Insufficient source input is flagged.
  • Valid generated output passes validation.
  • Empty model output is rejected.
  • Insufficient generated output is flagged.

Continuous Integration

GitHub Actions installs the application’s dependencies, compiles the Python source, and runs the automated pytest suite when changes are proposed or merged into the production branch.

6 / 6 automated validator tests passing locally and through the CI workflow.

AI-Enabled Operations

Workflow Automation

Applying AI Beyond a Standalone Application

Approach

My AI work builds on operational, documentation, and process-design experience. I focus on identifying repeatable work where AI or automation can reduce manual effort without removing the controls needed for reliable business processes.

Applied Experience

  • AI-assisted technical authoring and content-development workflows.
  • Documentation automation and maintenance tracking.
  • Prompt and instruction design for structured technical outputs.
  • Human-review checkpoints for AI-generated content.
  • Operational workflows connecting technical data, documentation, and decision support.
  • Identification of repetitive processes suitable for automation.

Tools

OpenAI API GitHub Copilot Microsoft Copilot Writer.com Cursor Lucid AI Power Automate

Technical Capabilities

AI · Automation · Delivery

From Use Case to Production

What I Can Do

  • Identify business and operational use cases appropriate for AI assistance.
  • Translate workflow requirements into application behavior and structured prompts.
  • Build Python-based AI workflows using external model APIs.
  • Design grounding rules and human-review controls.
  • Test model behavior using repeatable regression scenarios.
  • Build automated application validation with pytest.
  • Configure GitHub Actions CI workflows.
  • Manage application secrets outside source control.
  • Deploy working applications from a version-controlled production branch.
  • Connect technical implementation to operational governance and business outcomes.
AI examples are designed using generalized operational scenarios and do not expose confidential company data, systems, or processes.