Because this is about product design, UX/UI workflows, understanding how people end up using systems, and software engineering…I LOVE LOVE LOVE Figma as a product. Always have. I’m long on Figma, and this is not investment advice since they IPO’d, but since I love Figma as a product and realize people might not be familiar with it unless you’ve used it before when working closely with SaaS Product teams, I’ll embed the Case Study as a Figma Make item!

Below is the live, interactive case study mapping the original 1.6M-record pipeline and 48-hour EMR bypass engine I engineered for Alameda County and scaled across 32+ cities.

NOTE: If the embed doesn’t render on your device, explore the live interactive system architecture here.
🙂

If I were tasked with solving this exact 1.6M-record pandemic chaos today, here is how I would split the workload between modern out-of-the-box AI tools and hands-on engineering:

• Automated Regex & Schema Generation: Use Copilot or Databricks Assistant to draft complex text-parsing functions for phone extensions and malformed emails in seconds rather than coding everything from scratch.
• Multilingual Unpacking & Translation: Deploy small, specialized LLM agents (or Azure OpenAI endpoints) to automatically tag language preferences (e.g., Farsi, Spanish, Mandarin, Tagalog) and translate unstructured free-text notes into standardized intake fields.
• Household Name Parsing: Use zero-shot entity extraction to instantly detect when a single cell contains three family members, automatically structuring them into parent/dependent JSON records.
• Automated Synthetic Test Data: Generate thousands of dirty edge-case CSV rows instantly to stress-test pipeline limits before touching live production data.

FIRST: You still need to know what you need to know! The reason why my solutions engineering worked is because I understood the edge cases all of the automated systems missed to begin with. When this is your perspective, you inherently know AI systems are going to behave similarly because they don’t know what they don’t know. That is why PEOPLE are magic and can never be replaced by AI.

• HIPAA & PII/PHI Boundary Enforcement: AI models cannot guarantee zero-leakage compliance on raw, unscrubbed free text. Strict data sanitization and local scrubbing boundaries must be hardcoded before data hits an LLM.
• Batch Chunking & Importer Limits: Modern AI cannot magically fix legacy platform memory caps. Hardcoded payload boundaries (like our 10,000-row × 30-column chunks) still require explicit software engineering. And knowing the limits of your system or your client data requires personable working client relationships because everything is nuanced and case-by-case.
• Suspect Queue QA & Field Coordination: When data is ambiguous, AI should never “guess.” Routing edge cases to an auditable QA queue for human case worker follow-up remains a non-negotiable rule of Restorative Digital Justice.
• EMR/EHR Business Logic: Negotiating data models and bypassing stalled vendor integrations requires high-trust stakeholder alignment with clinic directors, which is something no algorithm can automate.