So are you going to hit record and walk away?
Alright, here we are from 5 20 22, and we're here with Allister Blake with inference, Allister. Nice to meet you. Nice to meet you too. Thanks for the time. So that gives people an indication I have not met with you before. , I've had John Hall, I've gone to the show a number of times, and he works with you pretty closely.
And he said, you've got to talk to Alice, or you got to understand what inference is doing. So people are going to get a feel for this. This is what I used to do as a CIO. When I met a new company. Okay, we're going to go through the education process and friends. What do you guys do? Yeah, so
basically we're a company that unlocks the power of VHR data by using NLP models, , to really extract insights from unstructured data.
And we have the great privilege of working with John Halamka and colleagues at Mayo clinic to be their partner in extracting all of this rich insight from EHR data at Mayer
EHR data NLP. Those are the two words I heard. All right. So we have a ton of data locked up. Yeah. All sorts of, or yeah.
Unstructured data within the EHR. That's what you're working with. Absolutely. All right. So, I mean, don't, we already have that built into epic or into Cerner or into an Athena Meditech?
Well, it's absolutely about HR nature. It's also about other modalities of data too, and integrating those as well. So we can talk about that in a second, but to your question around, yes, there's all this information in epic.
Even within the structured fields, what you see is that there are some times 50 different ways that a single blood test is codafide instruction variables. And so you need to do a lot of processing of even the structured data to make it really easy to use. And then further than that, there's all that richness that the physicians write in all life, all this free text boxes in epic and et cetera, that is completely unleveraged by researchers and pharma companies and all the downstream uses of the.
are you helping to clean up that data? Are you actually taking that data out of its locked state and moving it into a, , discrete data elements that are, , more standardized that we can play with very
much end to end. So with mayor, we deploy our platform on top of their unified data platform, and then we process all of the data in there.
And then we also act as their commercialization arm to partner with third parties, such as the biopharma in life sciences world. We do that provisioning of the data in a very clean way. And then we also work with third parties.
Is there, is there any aspect of this that is about security or privacy?
So there's a huge element of that.
And there's two key ways that we address the kind of security and privacy, , aspects of, of using EHR data. The first is that our platform is deployed in our host AMCs, , cloud instance. So it may call this data by. So, , Mayo is deployed in their GCP, but with our next AMC partners that we're building a network of, it's going to be an Azure for one of them and an AWS for another.
so it's very flexible. It's agnostic cloud cloud agnostic, I guess,
is what the cloud agnostic exactly and data behind glass. So did and never leaving the host institution. So that's privacy component.
Number one, data never leaves the host institution. So you're putting this in the. But it is there it's a virtual private,
it's their own cloud.
So with Mayo, it's their own GCP cloud. And then the second element obviously is de-identification. So we're only really using the states at once. We've identified it. And that's the first step. And we've worked very closely with, , one of the industry leading statisticians to certify that the identification approach that we have and, , and.
Pleased about how high it forms both on the structured data and only unstructured data, because that's a crucial, , capabilities.
All right. So w so my house is, so we'll partner with someone who's doing research on this data. They are going to, , well, they, , you say that the data never leaves the, the plant.
This is the thing, I think that's the state. This is what I'm trying to get to it. So if the data never leaves. But I'm working with this third party that wants to do research on this data. How does that work? Yeah, it's
great. So the way we kind of go to market and what without third party pharma partners is in two ways.
One is that in this world of using real world evidence, many pharma companies want someone to help them do a lot of the analysis and work with the data for them. So we have a team of data scientists who access the data from my host AMCs and do the insight generation. Piece number one, but model number two, that we're really pushing for in the near future and about to sort of launch our, , our, our product in the next quarter is provisioning a workspaces environment whereby these third parties can access that platform behind the host institutions, , , within their environment.
So very much kind of provisioning, safe, secure access to the data behind glass within each institution.
What health systems come to you? The problems they're trying to solve, I assume that's one of the problems we're trying to solve, protecting their data whilst enabling third party collaboration.
Because the number of people, when I was CIO, the number of people that came to me and said, Hey, we need access to your data. We want your data. I mean, the, I mean, not to call anybody out, but IBM Watson was like, yeah, you're just going to give us all your data. I'm like, oh, wait a minute. I don't like that.
But a lot of, a lot of, , AI models and whatnot, that's how they think it's like, give us your data and we will provide you insights back. So that's one of the.
So one of the, so the use cases for AMC partners is that they get the researchers get to use this platform internally to support their research efforts, which is exactly how, , pharma partners use it as well.
So that's a huge use case for the AMC and to do so in a way that's privacy preserving as well as then enabling these third parties to safely access the data without having to do the, what you described of shipping out data and all this security and privacy
concerns. All right. Give me the other problem sets that people are coming to you and you go, okay.
Yeah, we, we solve that problem.
Yeah. So I think one of the big use cases that pharma see from the data from, , from these AMCs is how do I find insights about the efficacy of my products in the real world? So how do I know that my vaccine for COVID is superior to the other? How do I know that my drug is being used off label and it's actually got good outcomes for all those off label use cases.
Yeah. We see a range of different real world evidence requirements from the life sciences chick.
So where life sciences and the AMCs intersect you provide the we're
in between providing that technology and that partnership on both sides to be able to facilitate it being a really safe and, and, , value adding engagement
We talked about the EHR datasets, but what other data sets are we looking at here?
So we're increasingly looking to integrate imaging data. Particularly imaging and visual pathology. And we actually have a subsidiary company that's all about generating near digitized pathology data. And then secondly, genomic data.
So we know that the aspiration should be to go from genotype three to phenotype. So can we provision access to data? That's rolled away from the role fast queue and bam files for, for genomic data right away. Annotations from the clinical, the HR record.
That's fascinating. , AMCs, primarily that's the market you're looking at and working with pharmas, the primary
market for the customer.
Yeah. So really the primary kind of AMC partners are the kind of leading research institutions. We know that. One of the key things that's unique about AMCs as opposed to broader health systems is that they have a really big research component to their work internal and external, internal and external.
So they work a lot with the life sciences industry, but they also just have a lot of internal research efforts. And we know that as part of the partnership and providing this platform and then for them to use the data is a huge need for them. Otherwise it's the kind of developing a research platform is the kind of thing that is a CIO is often a cost.
Absolutely for, for a partnership with us, we very much like to enable our AMC partners to use the
platform themselves. So where's inference in terms of its life cycle as a, as a
company. So we've raised us series C about a year or so ago. We, , I've gotten great venture backing from matrix capital partners, as well as Mayo clinic ventures and entity ventures with.
Looking to really build, , AMC network of collaborators. So I've talked a lot about our partnership with the Mayo clinic, congenital LAMCO, but where this year, hopefully about to announce another two to three top AMC partners, sort of in the peer of the Mayo clinic sort of partners. And, and then hopefully go on from there with our,
when somebody says platform, when I hear platform and vagina, I fucked.
When I hear platform though, it's the more entities that use it, the more utility, the more value gets created out of it, but it almost sounds like each one of these want to be distinct entities. I mean, what aspect of this is a platform?
So it's, it's very much a sort of federated architecture and a federated network.
So that is a key element to this. How is it a platform question is that you can, if you have deployments in each of these AMC partners, you can derive aggregated. From across them and sort of platform, but also you can do so without having to extract any of the data. So this is the key,
that's an extra reason as a former partner, I can now go to three to four AMCs and look at the various things.
It's one thing I was just talking to somebody about, , big data and the problem with those massive data sets is the, , you know, the massive data set in the Northeast. It's not indicative of the Southeast. It's not indicative even of the ed. From 10 o'clock at night to seven o'clock in the morning. I mean that the data sets do matter.
And a lot of the datasets are, I don't want to say no, I will say local they're local. I mean, it's, they're, they're based on the geography, the weather that, I mean, how people interact with each other and that kind of stuff. And so I would have, actually, these pharma companies are trying to get very specific, , in their research.
I mean, obviously there's value in that the larger, just broad dataset that you would get from say a true vet or something like that. Significant value. I would think from those, those distinct data sets
in those markets. Right. And what we see with AMC data in particular is that for very specific requests of, I need these patients with this disease at this stage, typically they're very small numbers and often only found in these leading institutions.
And so that's a huge need for the. The life sciences industry. And that's why we're really focusing on building a network of eight.
So here's my catchall, because again, I'm, I'm just coming up to speed on this, on what you're doing. , what's the question I didn't ask that I should've asked that, you know, maybe I didn't pick up on something, a key aspect here.
great question. I think the, I think the question though, that the key theme that I would sort of flag is around the technology. I think one of the distinctive things about what we do is that. Very excited about our NLP and that is really the crucial differentiator. Okay.
How's it different than, I mean, NLP has been around for awhile for awhile.
course it is. I think the, the key thing is that many folks are just starting to only use this on unstructured data in the EHR and the kind of that really thinking about validating it without thinking. Having models that are trained or multiple different data sets that can really be robust in terms of identifying the extreme discrete entities from this unstructured data.
So I think that's one of the key things that we're thinking about is not just developing models, but trying to make them validated and make sure that it's not just a random output that we say is good, but one that is sort of thoroughly validated.
Fantastic. Well, I want to thank you, 📍 Alice, sir, for your time again, and, and bring me up to speed.
This is the education of bill Russell. I appreciate it. Thank you.