How this PM streamlines 60k-page FDA submissions with Claude, Streamlit, and clever AI workflows
Claire Vo (host), Prerna Kaul (guest)
In this episode of How I AI, featuring Claire Vo and Prerna Kaul, How this PM streamlines 60k-page FDA submissions with Claude, Streamlit, and clever AI workflows explores aI workflows cut months from 60k-page FDA submission drafting time The episode shows how life-sciences regulatory submissions (e.g., a ~60,000-page BLA/CTD package) can be accelerated using a GenAI-driven workflow that produces strict XML outputs and redacts PHI from messy clinical notes.
AI workflows cut months from 60k-page FDA submission drafting time
The episode shows how life-sciences regulatory submissions (e.g., a ~60,000-page BLA/CTD package) can be accelerated using a GenAI-driven workflow that produces strict XML outputs and redacts PHI from messy clinical notes.
Prerna starts by prompting Claude like a “software engineer,” using a PM-style narrative (impact, stakeholders, demo goals) to get a full implementation plan, setup guide, and code scaffolding.
She then wraps the workflow in a Streamlit UI for non-technical users, adds synthetic-data generation for testing, and includes trace-and-cost transparency to address organizational concerns about scaling AI usage.
In a second workflow, she uses Anthropic’s Prompt Generator and Projects to create a structured stakeholder-influence coach, grounded in public-domain persuasion/literature sources, that outputs meeting prep, agendas, and anticipated exec questions.
Key Takeaways
Treat the LLM like an engineering partner, not just a chatbot.
Prerna’s strongest results came from giving Claude a PM-grade problem statement (why it matters, expected product, demo goals), which yielded not only code but also setup docs and a demo narrative—accelerating initial solution shaping.
Get the full analysis with uListen AI
Strict output formats (XML) must be a first-class requirement.
Because CTD/BLA submissions require rigid XML structure, she made “structured, schema-like generation” a core constraint from the start—reducing downstream rework and making the workflow viable for regulated submissions.
Get the full analysis with uListen AI
PHI redaction is harder in free-text notes than in tables.
Clinical data contains both structured fields and unstructured clinician notes; effective redaction needs medical NER-style detection beyond column-based masking, especially to catch names, dates, and other identifiers embedded in prose.
Get the full analysis with uListen AI
A simple UI can be the adoption unlock for non-technical stakeholders.
Wrapping the pipeline in Streamlit turned a script into a usable internal tool—critical for cross-functional teams like medical writers and regulatory specialists who need buttons and previews, not notebooks.
Get the full analysis with uListen AI
Use synthetic data to test privacy and formatting workflows safely.
Generating synthetic trial data enabled validating PHI detection/redaction and document outputs without exposing real patient information—an important practice in regulated environments.
Get the full analysis with uListen AI
Make AI costs legible to overcome internal resistance.
By instrumenting per-operation cost and duration (e. ...
Get the full analysis with uListen AI
You can operationalize ‘soft skills’ with structured prompting and curated knowledge.
Her stakeholder-communication coach uses structured, XML-like prompts plus a knowledge base (e. ...
Get the full analysis with uListen AI
Notable Quotes
“We had to develop a nearly 60,000-page document. Would have taken about four to six months of effort and nearly 20 specialists...”
— Prerna Kaul
“The thinking I had in mind is that Claude is a software engineer... trying to tell them why it matters... what end product we want to produce.”
— Prerna Kaul
“The first is... the BLA is a structured document... XML-based... The second was... detect PHI... because it's patient data.”
— Prerna Kaul
“If you're saving on time, you're bringing life-saving vaccines in the hands of people who actually need it.”
— Prerna Kaul
“If you're getting internal resistance to cost... bring true transparency [to] ROI and investment.”
— Claire Vo
Questions Answered in This Episode
On the FDA submission workflow, what specific CTD modules did you generate automatically, and which parts still required human medical-writer judgment?
The episode shows how life-sciences regulatory submissions (e. ...
Get the full analysis with uListen AI
How did you validate PHI redaction quality (precision/recall) on unstructured notes, and what was the human review process before anything could be used in a real submission?
Prerna starts by prompting Claude like a “software engineer,” using a PM-style narrative (impact, stakeholders, demo goals) to get a full implementation plan, setup guide, and code scaffolding.
Get the full analysis with uListen AI
What guardrails did you implement to ensure the XML always conformed to required schemas (e.g., retries, validators, constrained decoding, post-processing)?
She then wraps the workflow in a Streamlit UI for non-technical users, adds synthetic-data generation for testing, and includes trace-and-cost transparency to address organizational concerns about scaling AI usage.
Get the full analysis with uListen AI
Claude ‘searched up’ the CTD format—how did you prevent hallucinated or outdated regulatory requirements from entering the pipeline?
In a second workflow, she uses Anthropic’s Prompt Generator and Projects to create a structured stakeholder-influence coach, grounded in public-domain persuasion/literature sources, that outputs meeting prep, agendas, and anticipated exec questions.
Get the full analysis with uListen AI
What did stakeholders (regulatory, clinical, legal) push back on most: accuracy risk, auditability, or model/vendor choice—and how did you address each?
Get the full analysis with uListen AI
Transcript Preview
So you are working in the life sciences on really health-impacting vaccines and treatments, and with all that amazing investment and scientific work also comes a lot of paperwork.
We had to develop a nearly 60,000-page document. Would have taken about four to six months of effort and nearly 20 specialists, not to mention the millions of dollars spent.
Where did you start?
I gave Claude the problem statement, the pitch, and a demo. The thinking I had in mind is that Claude is a software engineer, and I'm talking to them and trying to tell them why it matters, like any good PM would, and trying to tell them what end product we want to produce as a result. And not only did it fully understand what I was asking it to develop, it created a little setup instructions markdown file for me, and it gives me all of the setup instructions and all of the tasks that it does and its capabilities, which is super handy.
Were you really able to shift the cost there? Did you see an impact here?
We did. We did. And any kind of cost savings you can generate have a direct impact to the bottom line. And in addition, if you're saving on time, you're bringing life-saving vaccines in the hands of people who actually need it.
[upbeat music] Welcome back to How I AI. I'm Claire Vo, product leader and AI obsessive, here on a mission to help you build better with these new tools. Today, we have Prerna Kaul, who's taken years of machine learning experience and helped develop some of the products you know and rely on every day at companies like Amazon Alexa, Moderna, and Panasonic. She's going to show us how she uses AI to accelerate drug discovery and vaccine approval, something that takes tens of thousands of pages of documents, dozens of people, and months and months and months of work. She'll also show us how you can use Jane Austen, Dale Carnegie, and some AI to manage your stakeholders just a little bit easier. Let's get to it. This episode is brought to you by CodeRabbit, the AI code review platform, transforming how engineering teams ship faster with AI without sacrificing code quality. Quality code reviews are critical but time-consuming. CodeRabbit acts as your AI copilot, providing instant code review comments and potential impacts of every pull request. Beyond just flagging issues, CodeRabbit provides one-click fix suggestions and lets you define custom code quality rules using AST grep patterns, catching subtle issues that traditional static analysis tools might miss. CodeRabbit brings AI-powered code reviews directly into VS Code, Cursor, and Windsurf. CodeRabbit has so far reviewed more than ten million PRs, been installed on one million repositories, and has been used by 70,000 open source projects. Get CodeRabbit free for an entire year at coderabbit.ai, and use the code HOWIAI. Prerna, it's so nice to have you. I'm really excited about your workflows.
Thank you so much for having me. Very excited to be here.
Install uListen to search the full transcript and get AI-powered insights
Get Full TranscriptGet more from every podcast
AI summaries, searchable transcripts, and fact-checking. Free forever.
Add to Chrome