How Prerna Kaul Automates 60,000-Page FDA Submissions and Coaches PMs with AI

By: Prerna Kaul Host: Claire Vo Source: How I AI (ChatPRD) Type: podcast

Summary

How I AI episode with Prerna Kaul, product and platform leader (Amazon Alexa, Moderna, Panasonic Well) working in life sciences where regulatory submissions can reach 60,000 pages. Two workflows solving technical and human challenges: (1) Automating FDA Biologics License Application (BLA) documents — treated Claude as a co-founder and software engineer, gave a high-level prompt with problem description, business impact, and desired outcome; Claude generated a complete project plan in minutes (Markdown instructions, capabilities list, demo narrative script, Python starter code) vs. a week of traditional PM + engineering planning; built PHI (Protected Health Information) redaction using medical named entity recognition models and complex RegEx pattern matching; wrapped everything in a Streamlit UI so non-technical colleagues could generate synthetic clinical data, detect/redact PHI, and produce the Common Technical Document (CTD) with both text and proprietary XML download formats; built-in cost analysis feature tracks token costs and duration per operation for transparent ROI comparisons against manual review; (2) AI-powered communication coach for stakeholder management — used Anthropic Console Prompt Generator to create a structured master prompt for an “Influence and Communication Coach” with XML-like format; uploaded classic literature from Project Gutenberg (Dale Carnegie’s “How to Win Friends and Influence People,” Jane Austen works, persuasion books) as knowledge base in Claude Projects; generates realistic high-stakes scenarios with competing stakeholders (CEO, Head of Sales, Head of Clinical Data, Chief Legal Officer); produces comprehensive strategic playbooks including situation analysis, tech leader perspectives (Nadella, Jassy), Carnegie communication principles, day-by-day pre-work plans with specific stakeholder conversations, minute-by-minute meeting agendas, and back-pocket answers for curveball questions.

Key Ideas Extracted

Claude as co-founder, not just code assistant: Treat Claude like a software engineer — explain why the problem matters, the business impact, and desired outcome; it generates complete project plans, not just code snippets
PHI redaction without specialized data scientists: Claude identified correct medical named entity recognition models and generated complex RegEx for pattern matching with minimal domain-specific instruction — work that normally requires months of model development
Streamlit for democratizing AI tools: Wrap Python scripts in Streamlit UI so non-technical colleagues can use sophisticated AI tools directly — runs from terminal with streamlit run app.py
Built-in cost analysis for stakeholder buy-in: Track token costs per operation (e.g., PHI redaction cost per patient) and build transparent ROI comparisons directly into the tool — preempts the “AI is too expensive” objection
Anthropic Console Prompt Generator: Use the built-in tool to create sophisticated, structured master prompts with XML-like format — generates better prompts than manual writing for complex coaching scenarios
Classic literature as AI knowledge base: Upload timeless works on persuasion and human dynamics (Carnegie, Austen) from Project Gutenberg to ground communication advice in proven principles
Day-by-day strategic playbooks: AI coach generates not just high-level advice but specific pre-work plans, stakeholder-by-stakeholder conversation guides, minute-by-minute meeting agendas, and prepared answers for anticipated pushback
Problem-first approach to AI adoption: Define the most painful, time-consuming bottleneck first — whether technical (60,000-page documents) or human (stakeholder alignment) — then architect an AI solution around it

Notes

Published Jul 14, 2025 on How I AI (ChatPRD). ~9 min read. (Note: filename says 2025-09-08 but actual publication date is Jul 14, 2025)
Sponsors: CodeRabbit, Lovable
Prerna Kaul background: Product/platform leader; Amazon Alexa → Moderna → Panasonic Well; works in life sciences/regulatory
Tools: Claude (planning + coding), Streamlit (UI), Python, Anthropic Console Prompt Generator, Claude Projects, Project Gutenberg (knowledge base source)
Domain context: Biologics License Application (BLA) — can reach 60,000 pages, 20 specialists, 6 months, millions of dollars
Key technical details: PHI redaction, medical NER models, RegEx pattern matching, CTD module generation, proprietary XML format
Two companion workflow guides published Jan 8, 2026
Cross-references: Regulated industry AI adoption, Claude as co-founder pattern, Streamlit prototyping, AI coaching/mentoring

Raw Content

Re-scraped from ChatPRD 2026-02-16. Full article content captured in Summary and Key Ideas above.