Selected work

Case study

OpenAI-Powered Documentation Quality Feedback System

Delivered privacy-aware AI coaching inside healthcare workflows to improve documentation quality, support clinical oversight, and make adoption measurable.

  • AI & Data
  • Healthcare IT

45%

fewer audit findings in the first quarter

Context

Overview

Delivered privacy-aware AI coaching inside healthcare workflows to improve documentation quality, support clinical oversight, and make adoption measurable.

The system ingested patient notes, removed PHI, and fed the content into OpenAI to score compliance against Blue Cross Blue Shield criteria. Feedback returned with severity rankings, remediation steps, and direct references to the relevant standards. Technician dashboards tracked progress, while compliance leads saw trendlines.

Security was non-negotiable: encryption at rest, strict access controls, and audit logs tied into existing governance. The result was a coaching loop that caught issues before audits did, shortened training cycles, and gave leadership measurable confidence in documentation quality.

Proof

Impact

  • Down 45%

    Audit exceptions

    First 90 days

  • 20+ hours/week

    Review time saved

    Automated triage

  • +40% message opens

    Feedback adoption

    Tone-tuned prompts

  • 30% faster

    Revision speed

    Strengths-first coaching

  • 45% fewer audit findings in the first quarter

  • Automated feedback loop scaled across 12 provider teams.

  • 30% faster revision turnaround thanks to balanced coaching.

  • Teams message engagement rose 40% after tone adjustments.

Highlights

What shipped

  • Built a secure pipeline that de-identified PHI before it entered the LLM workflow.

  • Delivered feedback within 30 seconds, prioritizing severity and remediation guidance.

  • Integrated with existing EHR workflows so technicians saw coaching inside their daily tooling.

  • Crafted role-specific prompts so technicians, providers, and compliance leads received tailored guidance.

How it unfolded

Journey

  1. **Field discovery:** Shadowed technicians to observe note-writing friction and ran lightning interviews about coaching needs.

  2. **Secure data plumbing:** De-identified PHI via Python + SQL pipelines before passing notes to the OpenAI API.

  3. **Prompt co-design:** Co-created blueprints with auditors and BCBAs so the model assessed medical necessity, clarity, and compliance.

  4. **Tone iteration:** Ran rapid feedback loops to soften language, spotlight wins, and keep morale high.

  5. **Operational launch:** Embedded coaching inside Teams alerts, dashboards, and compliance workflows.

What stuck

Lessons

  • Empathy scales adoption

    AI feedback lands when tone and framing feel supportive, not punitive.

  • Humans plus AI win

    LLM insights paired with clinical oversight created better coaching than either alone.

  • Measure and iterate

    Tracking opens, revisions, and audit outcomes kept the system improving week over week.

Personal note

Reflection

Documentation quality became a coaching conversation instead of an audit firefight—freeing technicians to learn faster and giving leadership real-time confidence in compliance.
Continue the conversation

Next steps

Need AI layers in your clinical workflows?

Let's explore how LLMs, structured templates, and audit trails can fit into your charting, intake, or coaching operations without introducing hallucination risk.