Abhishek and Eric Ma on LLM-generated acceptance criteria for regulated industries
Recorded 2026-09-15 at 1531 via Plaud. 04:23.
Summary
A four-minute conversation recorded at the GxP and AI conference Eric is attending this week, immediately after a talk. Speaker 1 is Abhishek, who gave the talk; Eric Ma asks the questions. Abhishek opens by noting that a six-month MIT class he took was already six months behind in what it taught about LLMs: the academic view is worth having, but LLMs change so rapidly that coursework ages fast, which shapes his search for someone who can actually guide him. Eric remarks that Abhishek’s slides carried imprints of leading practitioners and asks whether he follows them; Abhishek says he keeps almost everything public information, with a couple of pieces taken from elsewhere.
The substantive core is how Abhishek produced a novel acceptance criteria statement for regulated-industry documentation. He built a system that ingested all the regulatory documents he works with, then spent a long time with an LLM prompting it to write acceptance criteria grounded only in what he had done for his own applications. The model returned a basic statement, which he instructed it to generalize, and he then double-checked originality by searching whether anyone else had said it: nobody had, while the statement remains valid against the regulatory corpus. Eric, who moved from unregulated research into regulated digital and data science work, names the pain point this solves: facing all the documents people have to write in a regulated environment, how do you even start, and templated statements help. Abhishek plans to copyright the statement and post it on LinkedIn. Eric calls it the most practical talk he heard that day, says he will quote it as “Abhishek said this”, and has sent Abhishek a LinkedIn connection request.
Plaud AI notes
App-generated, verbatim from Plaud’s Summary tab.
Core Synopsis
A specific, LLM-generated acceptance criteria statement is the central asset emerging from this conversation. The creator, Abhishek, developed this statement by ingesting a corpus of regulatory documents into a proprietary system and then using an LLM to generate and generalize a novel framework based on his own application work. The resulting statement is perceived as highly pragmatic and valuable, particularly for navigating regulated environments where documentation is a significant bottleneck. The creator confirmed its originality through search and is now considering copyrighting the asset before publishing it on LinkedIn to establish ownership and share it with a wider audience.
The Genesis of a Reusable Asset
1. The Identified Pain Point
An attendee, operating in the regulated digital and data science space, highlighted a common and significant challenge: the difficulty of starting documentation in a compliant manner. The regulated environment demands extensive, precise documentation, creating a “How do you even start?” problem for practitioners accustomed to less formal research settings. This created the context for the value of a pre-defined, pragmatic template.
2. The Solution’s Development Process
The creator, Abhishek, detailed the systematic process used to generate the novel acceptance criteria statement:
- Corpus Ingestion: A proprietary system was fed with a comprehensive set of regulatory documents.
- Initial Generation: An LLM was prompted to generate an initial acceptance criteria statement, specifically tailored to the creator’s own application examples.
- Generalization: The LLM was then instructed to generalize the initial statement, making it broadly applicable beyond the specific seed examples.
- Originality Verification: The creator used search to confirm that the generalized statement was novel and had not been previously published. The system confirmed the statement was a valid, original synthesis based on the provided corpus.
3. The Path to Distribution
The high practical value of the statement prompted a discussion on its dissemination. The creator intends to formalize ownership before public release.
- Intellectual Property: The creator is considering securing a copyright for the statement to protect his intellectual work.
- Publication Channel: The intended platform for sharing the asset is LinkedIn, which will allow for professional attribution and broad visibility.
Next Move
@Abhishek:
- Secure copyright for the generalized acceptance criteria statement - [TBD]
- Publish the copyrighted statement on LinkedIn - [TBD]
Transcript
[00:09] Speaker 1: Spending time on that.
[00:11] Eric Ma: Same here.
[00:13] Speaker 1: I took a six-month class from MIT.
[00:21] Eric Ma: MIT.
[00:22] Speaker 1: Cool. But even they are six months behind in what they are teaching. Yep. And LLM is changing so rapidly. I’m like, okay, it’s good. Good to have that view, vision of academic view. But what the LLMs can do and all, right? So they can. But that’s a good question. Is when I’m looking for somebody who can kind of guide in their life.
[00:54] Eric Ma: I was looking at your slides, and it had a lot of the imprints of some of the leading practitioners. I was curious whether you followed them as well.
[01:08] Speaker 1: Yeah. So if somebody is, I try to keep almost everything public information. Yeah, couple of places I have taken stuff, but those are in the box. Ah, yeah. But how how we can connect? I’m thinking of posting this some of the like the acceptance criteria statement that I came up with. Yeah, so. I’m gonna put some copy right and put on my LinkedIn.
[01:37] Eric Ma: Yeah, that’s a good one. That’s a good one. Very pragmatic. I definitely appreciate that. Yeah, because I’m I’m used to the regulated regulated space. Okay. And looking at all the documents that people have to write, yeah. How do you even start?
[01:52] Speaker 1: You’re in IT. You’re in IT or QS?
[01:54] Eric Ma: I’m in digital. You’re in digital. Okay. Data science. Okay. And previously, only in research. Okay. Which is unregulated, whatever you need to solve the research problem. But it’s different in the regulated environment. Yeah. So I thought having some templated statements were actually very helpful. Yeah. How did you How did you come up with that? I’m curious.
[02:22] Speaker 1: So. So I. So the system that I created. Yeah. It. I have ingested all the regulatory documents. Okay. All the regulatory documents are there. Then I played around with LLM for a long time. Yeah. That I want to come up with a statement for acceptance criteria only based on what I have done for my applications.
[02:55] Eric Ma: Huh.
[02:56] Speaker 1: Huh. Using these regulatory. So it gave me a basic statement. Then I said, “generalize this”
[03:06] Eric Ma: Generalize.”.
[03:09] Speaker 1: Then I even double check. Yeah. Has anybody else said this? Yeah. It search like no, nobody else has said this. Right. Right. But it’s it’s it’s valid. Yeah. Based on your corpus. Yep. But it’s not. It’s not that anybody else has. Cool. So I was thinking about doing a copyright.
[03:34] Eric Ma: Cool. That’s cool. That’s cool. That’s a good one. Yeah. Very pragmatic. I can like, basically, now take quote it. You know, like, guys, Abhishek said this. A guy I know from a conference. Yeah. Name is Abhishek. He said this. It makes a ton of sense. Yeah. You gotta write it the same way. Cool. Yeah. Thanks for that talk. Yeah. Out of the many talks that I’ve heard today, like this one is the most practical that I’ve seen. It is great. Like and subscribe. Yeah. We’ll do. I did send you a link to the connection request, so it’s you’ll find it in your inbox there. Thank you. Thanks a lot, Abhishek. Likewise.
Related
- Eric’s Field Notes MOC
- Home
- Validating an AI system in a regulated industry means validating a measurement system - how the novel statement was generated and originality-checked
- Academia lags industry by years in teachable practice - Abhishek’s MIT class six months behind on LLMs