Agentic AI and Automation

Capability portfolio

Agentic AI & Automation

I build human-reviewed AI systems around real work. The goal is controlled acceleration, clear evidence, and useful decisions.

AWS documentation produced through an AI-assisted validation workflow

How I approach agentic systems

Start with a bounded decision

I define what the system should collect, evaluate, recommend, and leave to the human. That keeps automation useful without pretending uncertainty does not exist.

Make evidence visible

Outputs should show their inputs, confidence, exclusions, and unresolved questions so the reviewer can make an informed choice.

Build feedback loops

I use validation, test data, history, cache behavior, and human review to improve the system without losing traceability.

Design the operating experience

The interface, documentation, safety boundaries, and recovery path matter as much as the model or prompt.

Job Surge

Designed a local-first, human-reviewed job discovery system that searches supported job sources, verifies active listings, compares them with experience and application history, and produces a decision-ready dashboard.

  • Adaptive search cadence and cache protection
  • Evidence-based scoring and exclusion reasons
  • Verified employer or ATS routing
  • Usage and cost tracking
  • Resume, cover letter, and application-answer workshop
  • No automatic applications, messages, or LinkedIn sign-in

The repository and personal data remain private. Portfolio visuals show the workflow without exposing API keys, search history, or proprietary configuration.

C2C Pattern Pack Builder

Designed an end-to-end production workflow that converts a one-pixel pattern source into written crochet instructions, countable graphs, branded documents, protected PDFs, listing imagery, quality checks, and a packaged deliverable.

  • Source-of-truth image validation
  • Adaptive row and color processing
  • Automated DOCX and PDF generation
  • Multipage graphs with coordinate support
  • Render validation and packaging checks
  • Workflow design spanning browser automation and a local application

AI-assisted documentation at AWS

Used Q CLI and MCP-connected documentation guidance to generate CLI examples, validate JSON, identify drift, and fact-check content during the Asset Model Interfaces launch. The hybrid workflow reduced an expected two-week process to three days. The second review resulted in only ten minor edits.

I also created a prompt-engineering course used by 45 writers during a four-week AI enablement initiative for a 150-person documentation organization.

AI recommendation workflow at Meta

Built an AI-powered recommendation workflow and interactive wiki experience that helped employees identify the correct datasets and tools. The workflow was estimated to reduce manual research by three hours per user. Internal details remain confidential, so the portfolio focuses on the information architecture, decision logic, and human-review model.

Tools and methods

Q CLI • AI agents • MCPs • Python • JavaScript • structured prompts • validation rules • local-first workflows • human review