Brainforge Legal AI Walkthrough ⚖️

Hey, recording this video just to do a quick walkthrough, just intro brainforge.

I put together some slides that I thought would be relevant for you.

So, you know, just to recap a little bit of what I heard over the weekend when we had, just met at the conference.

this.

This was basically just a rip off of the slides that you had already put together.

I thought it was a great way to frame kind of how, lawyers and maybe legal tech people kind of think about how AI fits across legal practice today.

some of these examples are just pulled from the transcripts from the session.

I know that we're generalizing it to more than just immigration, so.

But just at least wanted to put some meat where the bones were.

say.

You know, there is drafting and documents, so there is kind of this part, where there are all types of forms the types of documents that lawyers have to put together.

you know, case and workflow management.

This is the project management side of legal work where there are a lot of handoffs.

Collaboration is important.

There's all types of timelines to keep track of, prioritization.

yeah, there's document evidence review.

You know, this was talked about extensively at, the conference.

And then also, you know, research monitoring and client comps.

And so I kind of just sat there and I was reflecting on, you know, what are all the ways that brainforge has either worked with clients before in the legal space, what.

What do we actually think that we could do?

So I think this is a good way to frame it.

from.

From our perspective, you know, like legal AI, you know, it should be way, way beyond what, you know, a lot of what I'm seeing in the market, which is just like, you know, interacting with forums.

you know, the legal world doesn't need another Lagoa or Harvey.

I think there are enough of these, like, agentic surfaces where there's collaboration.

you know, It's the next version of a, Thompson REUTERS or LexisNexis.

So, but it's more than just about building the surface.

the.

Actually the things that move, the needle the most?

From brainforge's perspective, we believe that, you know, in a world where All models are kind of like, getting better and it's more of a level playing field, we've all gotten to the point where output quality is no longer, because the models are incapable.

It's really about how an organization manages context.

And so if I were to kind of like put my own spin on those four areas that you had laid out, this is how I would describe it.

So from a drafting documents perspective, one.

The number one most important thing is to define what does good look like.

I think the difference between a junior associate, And a senior partner is that the latter has had years of experience, decades of experience, has produced thousands of outputs, understands what good looks like, understands how to carry the firm standard.

and you want to build the structure around that.

Right.

and not only do you want to find like the, who's the kingpin, who's the benevolent dictator of the organization that really determines the standard for the organization.

I mean that's frankly just how Consultancies operates, how Brainforged operates as well.

but then once you get that standard, you need to be able to disseminate it across everyone else in the organization.

And that's where you build the harness.

And so there are many people who are using the same set of skills that want to drive consistent output.

Whether you're a junior associate or senior associate or kind of the managing director, partner, you want everybody to kind of have consistent output.

And that's the idea of building, using AI to help with that.

all, can all look all nice but like in order for it to be trustworthy, the inputs and the handoffs have to be very, very clearly, you know, documented and Yeah, that's, that's kind of I, how I see the first, the first bucket of work.

Then there's like kind of case and workflow management.

right.

So faster timelines.

You know, project management isn't just about kind of the surface that you use.

It's not just about Google Docs is not the only way that projects are to get done, but you know, in the engineering world we break everything down to tickets.

And so the ticket is the atomic unit for how work gets done.

we have project management software.

We talked about Jira.

so yeah, Jira is kind of having a renaissance moment right now because everything is around making the highest quality ticket and make it easy to hand off to an agent to run end to end.

and so internally we use Linear, because Linear we believe has the best kind of AI agentic support.

but obviously we deployed it in a JIRA environment as well.

then there's like intake prioritization.

Right.

So I think this is around, you know, well there's a, there's always like new cases and in a larger firm there's a bunch of cases.

Which ones do you take on first?

Right.

And so really understanding the prioritization framework for that firm, whether you're a high volume firm, and you're just cranking out cases, you're trying every, you treat every case the same, you're just trying to spend the least amount of time on each one or you know, you work on relatively fewer cases in a year and you know it's not first in, first out, deal, deal flow.

But you know, anytime a deal comes up, if there are certain parameters you need to be able to pull it forward and kind of the triage process is not, is not so it's not so linear.

and then really about building alerts and so AI allows us, when we have good tracing set up, that like anything around documenting consistency, risks, errors, all those things can be flagged before content is shipped and you can put it into teams, into Slack, wherever the firm is already working out of, making sure that humans are being able to set thresholds, understand kind of like what are, what's the risk we're willing to tolerate and when do we actually need to step in and take action?

That's super important.

Otherwise AI either kind of stops at every turn and then you're reviewing way too much and there's review overload.

That's what we're seeing on the engineering world or on the other side.

You know, you just let these cloud models just run or these cloud agents just run and there's not enough gatekeeping actually and they're wasting too much time burning so many tokens.

when you, you know the job should have been killed because of some risk.

yeah then there's like document and evidence review.

So document, evidence review I think about kind of extractions and summaries.

there are many different documents, thousands, sometimes hundreds of thousands of documents that in the legal world that need to be sifted through and you need to be able to pull out the most important points.

Right.

fortunately legal writing is quite structured.

if you read any case it kind of has the same elements there's the IRAC format and all these kind of legal writing frameworks so that even though the text is very dense, I think lawyers are really good at extracting out the most important pieces of content and that can totally be codified and just run through an LLM.

but then knowing kind of which pieces to surface, like how to actually do the triangulation, that's, that goes beyond what a generic LLM can do.

And that requires some evals where you're really understanding, like how does an attorney like think through, synthesizing different parts of a case?

You know, we talk about kind of how there are, there primary, primary cases or secondary cases.

you know, it's all part of like legal analysis, and being able to turn that into rules based, deterministic kind of logic, and then applying it to the props, making the models go through a lot more kind of review before they get to the output.

and so there's this whole world of kind of evals and observability that has sprung up where you can trace what is the fundamental unit of observability is around tracing.

And a trace can tell you everything you need to know about a particular job.

How many LLM calls that were made, whether the MCP was actually working properly, if there were any custom skills, how efficient was it.

there are so many more levers now that you can use to basically fine tune agent workflows.

It's a lot more deterministic than it used to be.

It's no longer kind of this black box that people think LLMs, typically operate out of.

and so when you're doing a lot of synthesis around text generation, summarization, these are the most expensive jobs.

And so how do you make sure that the right model is used per task and you're able to keep costs in a controlled manner?

that's going to be increasingly important for firms that are trying to adopt AI at a larger scale.

the last part is around research monitoring.

And so I'll kind of double down on like hey, there's like signal gathering, external signals, all the policy changes, any enforcement mechanisms that need to be learned, court changes for a particular jurisdiction that you're working at.

You know, it's hard for a lawyer to keep up with all this information.

You know, you can go to these conferences, you can read the news.

But you know, how are you actually taking what is, you know, the signal that you're getting from outside and decomposing it and bringing it into your process immediately.

Maybe there are certain forms that are no longer relevant and you need to shut them off and you need to spring up new workflows and there are new forms, there are new requirements there.

All of that is part of where AI can really push firms towards building this type of self learning loop that it will adapt as the policies continue to change.

then there's enrichment.

So I'm sure for your larger customers, larger clients, these firms have done thousands of cases.

There are all these signals that, that matter to them.

and you know, frankly there's a lot of cases where you're doing an additional personal investigation.

You're not just going to you know, take the case file at face, at face value, but you need to go and purchase data from you know, eDiscovery Services, and other kind of like third party data providers that are going help you enrich the data that you already have so that you can make the best judgments.

If you build a scoring model around kind of what, what's the case that has the most like highest likelihood of success, those are kind of things that you would need from like a very rich data model.

it's more of a machine learning challenge but AI definitely helps with the enrichment of it.

And then lastly is around comms, right?

So there's you know, you need to package the content that you're delivering into whether you're talking to an executive, an associate, different levels of detail.

This is where AI has already kind of made the most progress.

I no longer have to write any email from scratch.

Anything that I kind of like, even in this video as I'm recording, as I'm dumping out what we're working on, that transcript can be turned into different forms of messaging that are delivered to the stakeholder I want to reach.

So that's just a little bit more color of kind of.

I'd love to have this discussion in more depth of how rankforged can be helping legal practice.

I have a slide here specifically around you know, for Epic, you know, how I would consider, you know, designing a partnership.

here.

So here would be three opportunities.

I'm, I'm happy to kind of dive into this a bit more.

So the way I view this is like you know, Epic has 5,000 customers, they have all these relationships already.

you have kind of core service, which is like the ediscovery service.

You know, brainforge is kind of your FTE team.

We basically come alongside build, build solutions for your customers, to make sure that they're getting the most out of your services.

And also you can kind of how do you position your, your team, this, in this AI arms race, right.

and then the customer has the standard for what good looks like.

So I think there's a few low hanging fruit.

these are kind of concepts that we've kind of recycled across different industries.

You know, typically it's like, you know, something around intake, right?

Making sure that you're able to get qualified cases, and you're able to help triage them quickly because that's the number one revenue driver.

If you're able to build like a good system around that recommendation, engine or you know, just be able to speed up the way that you do intake, that is a huge, that's the biggest revenue driver.

then there's around kind of building scenarios.

And so yeah, with legal cases so much of it is about context gathering so you can understand the fact pattern fully and then you can build counterfactuals because you want to understand what are the other perspectives that I need to be able to defend against.

And so AI is really good at personifying different types of counterfactuals if you give it the right prompts.

And so I think there's very systematic ways to do this.

in, in the engineering world we're running sub agents all the time on top of our solution to basically black hat, white hat kind of our solution and to make sure that it clears a bar, in order to make it to production.

Then there's like file reconciliation.

So making sure that, hey, you know, the output, sometimes it's multi, threaded, making sure that you know, any alerts, signals that you need to, you know, flag to people because you know, there are certain, you know, there's white space.

You know.

Ultimately why AI is not effective is because humans are not effective at giving it the right data.

And so this is always going to be part of the handoff.

Some things get lost in translation.

And so you always need to be having like, you know, AI assisted reconciliation at every step of the handoff until one day where the AI can just hand it off to itself and you know they're going to be a lot better at doing that than humans are.

so yeah, I think those are a few ideas I'd love to dive into if any, if any of those seem interesting.

yeah.

And then the rest is kind of just like more slides about who brainforged is how we work.

you know, kind of this idea of bringing customers from Crawl, walk, run.

you know, we work with a wide range of customers across kind of their AI journey.

And so we understand, you know, how to identify where a customer is and then bring them to the next stage.

okay, I think this video is getting a bit long, so I think I'll just wrap it up there.

But hopefully this was a helpful just overview.

you know, if this is something that you'd be interested in discussing further, I'd love to book time with you.