0:07 hello everyone and welcome to the lth
0:09 product briefing I'm delighted today to
0:12 be joined by Jake weiner who is the head
0:13 of strategic business development at
0:16 Harvey Jake welcome thanks Nikki yeah
0:17 great to be speaking with you and
0:19 excited to show the Harvey platform a
0:20 little bit I think everyone is excited
0:24 to see this Harvey hard to believe given
0:27 people still feel that chat GPT and geni
0:29 and legal is recent but in fact Harvey's
0:32 now been around around for about 2 years
0:35 how has it evolved where is Harvey now
0:36 yeah probably the best way to answer
0:39 that is just to jump into the demo so
0:44 want to show Harvey's fault feature so
0:46 this is uh Harvey's sort of data room
0:50 tool this allows you to upload data sets
0:52 of thousands of documents and the
0:55 documents are then stored within Harvey
0:57 and you're able to then query across
1:00 them both in natural language as well
1:03 well as with dedicated purpose-built
1:06 workflows and tremendous tool for due
1:08 diligence transactional due diligence
1:10 for litigation Discovery and litigation
1:13 review tasks as well as for tracking
1:15 sort of precedents and deal points for
1:17 in-house teams particularly useful for
1:20 tracking agreements whether those are
1:21 commercial agreements or just agreements
1:23 with various sort of counterparties so
1:26 there a hugely valuable tool for working
1:28 with large legal data sets you to use
1:30 this you create a project and you can
1:33 see examples of some different projects
1:36 here and then you can jump into any one
1:38 of these and perform a query so let
1:41 start by just showing an example of one
1:42 of these so here we have a data set of a
1:46 number of commercial contracts and the
1:48 query that I I previously ran here very
1:50 simple query I just said for all these
1:52 commercial contracts tell me about the
1:54 parties any change of control Provisions
1:58 termination and red flags generally so
2:00 not particularly on this red flags point
2:01 a lot of specificity there so like
2:03 leveraging the model's knowledge to to
2:06 be able to produce this here we prompted
2:07 Harvey to produce this answer in a table
2:10 format and so you can see Harvey's gone
2:11 through every one of the agreements in
2:14 this data set it's produced this answer
2:17 in this table format specifying each of
2:19 those terms that we asked about we can
2:22 then open up any of these agreements and
2:25 we can jump directly to the terms that
2:26 we've asked about and see them
2:28 highlighted in the documents as well as
2:31 sort of Harvey Zone kind of analysis of
2:33 the terms that we asked about oh I see
2:34 there's red flags there that's
2:36 interesting are you able to customize
2:39 the data that is extracted from a batch
2:42 of documents so if you wanted say 25
2:44 very specific data points extracted you
2:46 can specify that absolutely and I can
2:48 show you exactly what that looks like
2:50 here's where you would run an open-ended
2:52 query like this you can have Harvey give
2:54 you an answer either in a table format
2:56 like we just showed or in more of a memo
2:59 format and so here if we want to get a
3:02 table you start by prompting in entirely
3:03 natural language and again you can ask
3:04 about anything you like I can say
3:08 something like tell me about the control
3:12 Provisions any supply chain risks
3:15 assignment and anything that would
3:18 implicate Georgia law right any sort of
3:19 like any term that you might happen to
3:22 care about and when we hit ask Harvey
3:23 you'll see that what it's actually going
3:26 to first do is take a a pass at
3:28 producing The Columns that that you're
3:30 likely to care about so you'll see here
3:32 it's a it's added a number of columns
3:34 related to two columns around change of
3:36 control you have the ability to edit
3:38 those so you can choose exactly what
3:40 each of these columns should should say
3:41 but again are there any provisions or
3:43 Clauses that implicate Georgia law what
3:45 are the details of those so so Harvey
3:47 gives you a sense of exactly what's
3:50 going to be in that table and entirely
3:52 flexible entirely open-ended and of
3:53 course this is powered by our large
3:55 language models so this is not using
3:57 conventional machine learning or pattern
3:59 matching it's not looking for Section
4:00 heading it's not even looking for
4:02 specific words it's actually reading the
4:04 context of the entire document to
4:06 produce these outputs and even once
4:09 you've even once Harvey's produced an
4:10 answer like this you also have the
4:12 ability to add columns right click this
4:14 button here we can add columns and add
4:16 files and then of course you have
4:18 sophisticated filtering tools so we can
4:21 filter these down and so great way to
4:24 like rapidly work through large data
4:26 sets of documents and perform these
4:29 sorts of diligence tasks okay the
4:31 example of prompting that I just showed
4:32 you is the way that you can use this
4:35 through open-ended natural language
4:38 prompting but as I mentioned we are also
4:40 building and releasing on an ongoing
4:44 basis sort of purpose-built endtoend
4:47 agentic workflows within Vault and also
4:49 throughout the Harvey platform generally
4:50 many of these are Under development
4:53 right now but within the assistant tool
4:55 within the various research tools there
4:57 are more and more sort of endtoend
5:00 workflows that don't require prompting
5:01 in the same way so you can see some
5:04 examples of these here here for instance
5:06 we have workflows provided around
5:09 analysis of Court opinions producing
5:11 email chronologies in for instance a
5:14 litigation context reviewing lease
5:16 agreements reviewing share purchase
5:18 agreements if we select one of these
5:21 create a do an LPA review or fund review
5:24 we click this Harvey specifies okay
5:25 these are the terms that you're likely
5:27 to care about you also have the ability
5:29 of course to edit any of these particular
5:30 particular
5:32 Concepts but it's setting it all up for
5:34 you you choose the documents within the
5:35 data set that you want to run this
5:38 across and then you essentially hit go
5:40 and we won't wait here while this runs
5:42 but this sort of gives you a sense of
5:44 where the Vault platform is going and
5:46 also where the Harvey platform more
5:48 generally it's going it's actually
5:49 running here but where the Harvey
5:51 platform is going more generally which
5:53 is more and more of these agentic
5:55 workflows that sort of remove the
5:57 prompting burden from the user and that
6:00 are purpose built for specific legal
6:02 tasks um conduct those tasks at really
6:04 high levels of fidelity large numbers of
6:07 these workflows are being loaded into
6:09 the Harvey platform and are just there
6:12 when you log in we also build these on a
6:15 custom basis for for customers and as
6:18 our models and LLS generally have gotten
6:20 more and more sophisticated we're able
6:23 to produce these workflows with really
6:25 pretty light engineering lift and we're
6:27 able to plug these in for customers on a
6:29 bespoke basis without a lot of obstacles
6:32 to to doing so and for those watching
6:34 who are not sure what the distinction
6:36 might be between prompting the assistant
6:39 tool or clicking a button on an agentic
6:41 workflow would it be correct to say
6:43 you're seeking a specific output you
6:45 click the button for that output and
6:48 then what Harvey does is take the
6:51 necessary tasks in order to achieve that
6:53 output or that outcome yeah that's right
6:55 that's right and I think the other way
6:58 to to think about it is many of the
7:00 tasks that that the workflows that we're
7:03 building accomplish are tasks that are
7:05 already possible in the Harvey platform
7:08 right now via open-ended prompting but
7:10 they might require very sophisticated
7:12 prompting or they might require like
7:15 multi-steps of prompting so running one
7:16 prompt taking that output then running
7:18 another prompt on top of it and then
7:20 doing that several times so what these
7:22 workflows are doing is essentially
7:24 putting together very sophisticated
7:26 prompts in addition to a software layer
7:28 um but very sophisticated prompts and
7:31 then chains of prompts so that as a user
7:33 you don't have to go through that that
7:35 process to get these sorts of outputs
7:37 got it thank you that's super helpful
7:41 great okay so next want to just mention
7:44 a few of these Harvey again like
7:46 workflow tools these are slightly
7:48 different so these are tools that again
7:50 are like purpose-built for specific
7:52 functions certainly won't dive into all
7:54 of these but tools for review of
7:56 litigation transcripts so trial
7:58 transcripts or deposition deposition
8:00 transcripts document comparison
8:03 substantive comparison of documents
8:04 whole document translation and then
8:06 certainly want to highlight this Redline
8:08 analysis tool so I'll jump into this
8:11 here this allows you to upload Redline
8:14 documents either PDF redlines or word
8:18 track changes and have our Harvey either
8:20 produce an issues list for you in the
8:22 pre-specified format again with really
8:25 no prompting or you can query this
8:28 generally here um some an amended set of
8:30 corporate bylaws and then for the prompt
8:33 just said you you represent a
8:34 prospective investor analyze these
8:37 changes flag any that might be of
8:39 concern from an investor standpoint and
8:41 you'll see that Harvey has done that
8:43 here it's extracted the changes that are
8:46 likely to be of concern for an investor
8:47 and then it's going to site back
8:50 directly to to the language within the
8:51 document here good example of sort of
8:54 the legal specificity of the Harvey
8:56 platform in terms of the ability to read
8:59 and analyze red lines okay great the
9:01 kind kind of like last point that I do
9:02 want to mention here just within the
9:03 Harvey platform before we jump to the
9:06 Microsoft Word ad in is the library
9:07 feature so we've talked quite a bit
9:09 about prompting and workflows and things
9:11 like that we are again building more and
9:13 more of these agentic workflows but
9:16 prompting and the flex ability and the
9:18 power that prompting offers with
9:20 Harvey's models is we think really
9:21 important and is always going to be a
9:23 part of the the platform and a lot of
9:25 folks are interested in understanding
9:26 how they can get better at prompting and
9:28 learn how to prompt and that sort of
9:30 thing we provide
9:33 and examples Library here we're
9:35 constantly updating this and you can
9:38 filter this across practice areas and
9:40 it's so great way to get a sense of how
9:42 to use the Harvey platform you can open
9:44 up any of these examples they include
9:46 sample documents and gives you a great
9:48 sense of how you can be using the
9:50 platform may not have thought about
9:52 using Harvey to produce
9:54 cross-examination questions in an expert
9:57 report but rvey will do that for you so
9:58 great way to get a sense of how to use
10:00 the platform and then we also do this at
10:03 the level of the actual interface so you
10:06 can load prompts and templates and you
10:08 can do this both using sort of Harvey's
10:10 own specified prompts and templates or
10:13 you can create prompts and templates
10:15 that are specific to your organization
10:17 or your practice area and so also I the
10:20 Harvey prompts the pre-loaded prompts
10:22 are such a great training tool actually
10:25 for people in how to actually prompt
10:27 properly and what good prompts might
10:29 look like I think that's right yeah yeah
10:31 you can see even the way that we lay
10:33 some of these out they're bracketed for
10:34 the user to be able to go in and edit
10:36 them a little bit so yeah they're you
10:37 can think of as helping you get to that
10:39 first step right they're like precedent
10:42 yeah so it does help you track usage and
10:44 so on in the platform from an
10:46 administrative perspective but let's
10:48 jump in and look at the at the word
10:50 addin yeah definitely yeah of course you
10:51 have the ability to track your history
10:52 and I'll just mention to like very
10:55 sophisticated controls from an admin
10:57 perspective in terms of data retention
10:59 ability of admins sort of review usage
11:01 statistics and data across the
11:04 organization we offer a client matter
11:06 number tracking and sort of
11:08 administrative tools for folks who need
11:10 to manage the use of AI tools for
11:12 specific client matters lots to be said
11:14 on like the administrative front but we
11:17 can save that for for a deeper dive okay
11:21 let me change my window here and we will
11:25 show the Microsoft Word addin so the
11:27 addin which you can see here on the
11:30 right hand pane allows you to language
11:33 directly in a Microsoft Word document so
11:36 the way that this works is hypo there
11:39 the way that this works is by letting
11:41 you select the language that you're
11:43 interested in editing and then you can
11:44 edit it in two different ways you can
11:47 either provide instructions to the
11:49 models or you can provide precedence we
11:51 think about this the same way that you
11:52 might give instructions to one of your
11:54 colleagues you might just tell them like
11:57 hey add a a particular term or make this
11:59 particular term more favorable to one
12:01 party or the other or you might actually
12:02 just give them a precedent and say hey
12:04 so you can see this kind of an action
12:05 here so I'm working in a credit
12:07 agreement I've selected the ma
12:09 definition starting with the instruction
12:11 based editing I can just say something
12:14 like make this more and you'll see that
12:16 Harvey's going to make suggestions on
12:19 the ways to revise this language you can
12:20 review these different suggestions and
12:22 see which ones you'd like to apply let's
12:25 see include an adverse effect resulting
12:27 from under any other material agreement
12:29 great we hit apply and Harvey's going to
12:31 go ahead and then make those changes in
12:35 track directly in the document again
12:37 this is powered by Harvey's legal
12:39 specific models so you'll notice that
12:42 these suggestions are I'm really quite
12:44 sophisticated and carefully tailored to
12:46 the content of what you're trying to
12:48 edit here so that's the way that the
12:51 instruction based editing works I'm
12:53 going to go ahead and just reject those
12:55 changes just so we have a blank slate
12:56 here and then I'll show the
12:58 precedent-based editing so we just click
13:00 over to precedent based here I've
13:03 uploaded another credit agreement as as
13:05 a precedent and we've just highlighted
13:07 this Mee definition which we want to
13:09 edit just hit make suggestions and
13:11 you'll see that what Harvey's doing here
13:13 is it's reading this precedent document
13:15 it's identifying the relevant language
13:17 in that precedent and then it's going to
13:19 make suggestions here based on the
13:21 precedent to bring this into Conformity
13:23 for instance specify that the financial
13:25 condition refers to financial condition
13:27 determined in accordance with gap great
13:29 we hit apply and Harvey will make those
13:32 edits in track changes here and again as
13:33 always you have the ability to view
13:36 references so you can go back and check
13:37 within the Press of where these
13:40 suggestions are coming from that's great
13:41 Jake thank you so much for showing us
13:43 all of this this is fantastic I think
13:45 for a lot of people would be really eye
13:47 opening just to see the breadth of
13:49 Harvey and the various problems that it
13:51 really can solve or the ways that it can
13:53 help lawyers I know that's really the
13:55 focus is on helping lawyers it really
13:57 feels like that so thank you so much for
13:59 sharing with us today it's great to hear
14:00 I'm an attorney myself we have many
14:02 attorneys here at Harvey as we've talked
14:04 about and so we're very passionate about
14:06 anything that we can do to help improve
14:07 the profession and improve the value
14:09 that the work that attorneys are doing
14:12 glad to hear that great we will put a
14:14 link at the bottom so anyone watching
14:16 can reach out to Jake and his team at
14:18 Harvey If you want to learn more Jake
14:20 thank you again and to everyone watching
14:22 thanks so much for coming along to the
14:24 LT product briefing and we'll see you
