Don’t buy snake oil
A Swedish legal AI startup called Legora has raised $550 million at a $5.5 billion valuation. This is absolutely unnecessary. This kind of global approach does not work because every court or adjudicating authority has its own format for writing applications, and every jurisdiction has its own laws and precedents. Any lawyer or firm works in particular areas of law. Therefore, any AI-based system will depend heavily on “personalization”: a basic structure that individual lawyers or firms can use to build their own library of legal articles, judgments, agents, and scaffolds for writing legal documents.
Can AI write complete documents?
Yes, but only if you provide it with organized information and break down the process step by step. Otherwise, it will write something totally useless, and you will waste a lot of time correcting it.
Introducing ‘Legal Eagle’ in Notion
I built my own system in Notion, which provides the world’s best tools for writing and productivity, including Gmail integration, calendar, collaboration, meetings, and billing.
Costs almost nothing & fully adaptable
I plan to release the templates for a $100 one-time cost. Notion charges $25 per month for its “Business Plan,” which includes all the features you need and more. You can use the latest versions of different LLMs in Notion, and they work just fine. There is no need for a specialized model for legal work, as Legora and others pretend.
I also plan to release a free version called “Legal Junior” (Notion’s $25/month charge will still apply) so people can get used to working with databases.
Currently, I have equipped it with scaffolds so it can write complete legal pleadings for the Calcutta High Court, as well as RTI applications and notices, with minimal prompting. The system is fully customizable. Scaffolds are easy to make. Notion AI can do it itself: just give it an example and ask it to make a scaffold.
Basic workflow (technicalities)
Step 1. Intake information
This is the manual part. A typical legal document will require many annexures (contracts, prior notices, correspondence, etc.). These are usually poorly scanned and may include stamps and handwriting. You will need to extract the relevant information using external tools. Grok is very good at reading messy PDF documents, and there are desktop applications like Normcap. You can also use Google Lens to extract text. If nothing works, type the information in once. Don’t be frustrated. Before OCR and AI, all this had to be done by hand.
Add these to the “Intake” database. Use Notion AI to automatically extract the relevant information.
Step 2. Organize the information
Add these (drag and drop) to the database that organizes case documents. Tag them as needed, for example, Notices and pleadings, issued or received. You can highlight and add your own comments (notes, analysis, arguments, etc.) directly on the document, which AI can also read.
Step 3. Write documents
Provide the basic arguments in plain language (fully express your frustrations), point to the annexures you want to use, and ask Notion AI to write your document. It will draft a complete document with the necessary dates, reference numbers, annexures, sections, tables, and prayers. Then make any necessary changes.
Step 4. Use it in court
All your case documents are organized in a database, so you can open what you need and search for information simply by prompting.
Legal Research
AI can search the web for important cases. Legal websites often publish articles about them, which you can import into the Notion “Articles” database and extract information from with a few clicks. However, for professional work, you need a subscription to a legal database, preferably one that provides email updates about newly uploaded judgments.
Legora claims it will support legal research and is setting up centers in various US states, presumably to build legal databases for local jurisdictions (one would think such web services already existed in the US) or to support customer service.
In India, such databases already exist. For example, I use Manupatra, which has its own AI-based and conventional search functions. You can download relevant judgments and import them into a Notion “Judgments” database. This lets you build your own library in the specific fields of law you practise, since no lawyer does everything. Manupatra also sends daily email updates, and Notion AI can search my emails too, so I can easily find recent cases with simple prompts.
You can also buy inexpensive legal books in India that contain detailed legal analysis and case law. Many also come in PDF versions that you can import into Notion. So, if you do not want to start building your library from scratch (which works too), you can begin with specialized books in your area (for example, constitutional law), and then keep updating your knowledge base.
Who can use it and how?
Any legal practitioner can use it, including lawyers, judges, and arbitrators. Provide your own scaffolds and it is ready to go. Beyond drafting, the system can track tasks, appearances, meetings and transcriptions, billing, and more.
How good is it?
It is very good, but you are still the master. Treat the agent like a super-assistant, and give very specific instructions. If you are vague, you are inviting mistakes. Having written many legal documents by hand, I can say this removes 80% of the work and frustration.
Have fun
I have worked on this system for a while (time flies) all by myself. I am offering it cheaply because I want more people to use it, especially in India. I want students and interns to use the AI-based system without having to think twice about the cost.