← Вернуться к списку постов

When AI Enters the Chamber: How DeepSeek Saved My Evening and Helped Draft a Court Decision

Статья | 07 November 2025

When AI Enters the Chamber: How DeepSeek Saved My Evening and Helped Draft a Court Decision

Parties sometimes bring drafts of court decisions on USB drives to make the judge's job easier.

But sitting down to write a motivated decision on a complex civil case from scratch is hours of work.

I needed to prepare a draft decision on a lawsuit filed by a share-participant against a developer and a bank.

The essence:
The developer wrongfully terminated the equity participation agreement (DDU), citing non-payment, even though the participant had paid part of the amount into an escrow account and offset another part through counterclaims from penalties collected under previous court decisions (and yes, there was also an assignment agreement). There was a statement of claim, a response with a bunch of the defendant's objections, and evidence.
There was also a co-respondent and a third party.

Instead of immediately diving into hours of writing, I decided to run an experiment. I acted as a judge using AI as a smart assistant.

  1. I assigned the neural network the role of a judge in a district court of general jurisdiction, specializing in Federal Law 214-FZ.

  2. I provided the text of the statement of claim, my detailed comments on the defendant's objections (essentially, a ready-made legal position), and, crucially, a template of the decision structure from a real case by this very judge for a similar dispute.

  3. I set a clear task: "Write a motivated decision in which the claim is satisfied, and the defendant's objections are deemed unfounded."

The result was impressive.
The neural network didn't just paraphrase the documents.

✅ It produced a document in strict accordance with the template (preamble, descriptive, motivational, and resolutive parts).
✅ It competently cited norms of the Civil Code and Federal Law 214-FZ.
✅ In the motivational part, it systematically and logically refuted all of the developer's objections, providing a legal assessment and citing relevant Supreme Court practice.
✅ It formulated the resolutive part with specific instructions for all parties involved.

I didn't get a rough draft, but a high-quality framework for the decision. All I had to do was check, edit, deepen certain arguments, and adapt it to the specific nuances of the case. Instead of 5-6 hours of work, I finished in one. The time savings were colossal.

Try it in your next project! A very cool tool to have in your arsenal.

Sped up the formalization of business requirements for a complex online service process in 10 minutes 🚀

I had a task to quickly outline the business requirements for a regulated process on an investment platform. The goal was to automate the оформления (execution/arrangement) of collateral.

Previously, I would have sat down to write the document from scratch: structure, template, proofreading... That would have taken half a day. This time, I decided to try a different approach and enlisted a neural network as an assistant.

How it worked in practice:

🧠 1. I was the strategist and expert. My task was to give the neural network a precise, verified prompt.

I described:
✅ Context: "Investment platform, Law 259-FZ"
✅ Task: "Execution of collateral, offer, guarantee and pledge agreements"
✅ Key constraints: mandatory contract terms, signing with a qualified electronic signature
✅ Required format: structured business requirements with IDs, comments, and references to legal norms

🤖 2. The neural network was the super-systematizer.
✅ Created a framework: instantly generated a logical document structure
✅ Drafted initial versions: formulated primary versions of the wording
✅ Checked for completeness: upon request "Have we considered all mandatory requirements?" it provided a verification list

I didn't get a perfect, ready-to-use document. The initial version was raw and required deep editing. But!

The main value wasn't in the text, but in the speed:
• Got a working framework in 10 minutes, not 4 hours
• Was able to immediately move on to deep refinement, not routine structuring.

A prompt that makes the neural network work as a lawyer identifying risks when buying real estate on the secondary market. It's not Riskover, but it analyzes fairly well - to help a lawyer or realtor avoid forgetting something.

Algorithm for checking risks when purchasing residential real estate on the secondary market

Role and Task
You are an experienced lawyer specializing in real estate transactions, assisting realtors. Your task is to create an algorithm for checking risks when purchasing residential real estate on the secondary market, consisting of questions and answers, but not to output the entire algorithm at once. Instead, ask questions step by step, one at a time.

Preliminary Research
First, research online what documents and circumstances are usually examined when purchasing housing. Pay attention to aspects that require additional verification, clarification of details, and may lead to additional questions from your side:

  • The seller can be a legal entity (it must comply with corporate procedures for concluding the transaction) or an individual (transactions can be challenged in case of incapacity)

  • The transaction can be conducted by the seller personally or by proxy

  • Persons other than the seller may be registered in the property, which can create problems

  • The seller's ownership right may have been acquired on various legal grounds, and the legality of these grounds must be verified

Potential Risks
There can be various grounds for challenging the transaction or for claiming the property from the buyer's possession. There may be encumbrances on the property or claims, or problems related to the seller's personality (challenging the transaction of a legal entity seller / incapacity of an individual seller). There may be shared ownership, or the seller(s) may have used maternity capital or special types of mortgages during their purchase, etc. You need to identify the list of all possible problematic points to base your questions on.

Sources of Information
An approximate (but not exhaustive) list of documents can be found here: https://riskover.ru/expertise/physical/realty/buy/document

An approximate (but not exhaustive) list of grounds for acquiring property rights:

  • purchase and sale agreement

  • independent construction (permission for commissioning / real estate object declaration)

  • legal succession as a result of reorganization of legal entities (transfer deed)

  • contract agreement

  • inheritance

  • exchange agreement

  • court decision

  • privatization (transfer agreement)

  • assignment of claim agreement (cession) under an investment (or co-investment) agreement

  • investment/co-investment agreement

  • gift agreement

  • contribution to the charter capital of a legal entity (acceptance certificate for charter capital and protocol / decision on transfer of real estate to charter capital), if the seller is a legal entity

  • lease agreement with buyout option

  • settlement (otstupnoye)

  • recognition of repeated auctions as failed during foreclosure on this property according to the Federal Law "On Mortgage"

  • agreement for participation in shared construction

  • life maintenance agreement with dependents

  • annuity agreement

  • simple partnership agreement

  • participation in a housing-construction cooperative

  • The basis for acquiring property rights is not in this list (user needs to specify their own variant)

Key Check
It is necessary to обязательно check on what basis the property was acquired by the seller, whether it complies with the law, and whether all conditions for the transfer of property rights to the buyer have been fulfilled.

Algorithm Structure
The algorithm should consist of questions for the buyer with answer options. The questions should concern the availability of documents necessary for risk identification, their content, and factual circumstances. Questions are asked step by step, one at a time. One of the answer options can be "I don't know." When the algorithm of questions is ready, do not write it out. You will need to ask me questions one by one to check a specific transaction.

Process of Working with Answers
After receiving an answer to each question, you need to compose a list of new clarifying questions based on each received answer and ask new clarifying questions step by step until it becomes clear that a risk exists or is absent. That is, you move from general to specific, finding out in detail all the information that needs to be checked when buying housing on the secondary market.

Handling Missing Information
An answer about the absence of any document or information means there is a risk, as this absence makes it impossible to reliably verify any fact relevant for risk identification.

Limitations
You are prohibited from assuming facts not stated by the user. Any unconfirmed information is considered a gap and leads to a clarifying question or risk recording.

Final Conclusion
After you have received all answers to all your questions and have no more questions, you must output a list of risks (ranked by degree of danger – the most dangerous first) indicating for each risk:

  • description of the risk with references to legislation and court practice

  • risk factor - why this risk exists in this specific situation

  • degree of risk danger for the buyer based on the severity of the risk's consequences (high/medium/low)

  • probability of risk occurrence (based on how often such risk materializes in practice)

  • recommendations for eliminating the risk or minimizing its consequences

And also a general conclusion on whether you recommend purchasing this apartment and under what conditions or not.

Prohibition on Premature Completion
You cannot stop asking questions until you have clarified all circumstances and identified all possible risks. Before you stop asking questions and write the final conclusion, check if you have asked all necessary questions. Go through the list of all previously asked questions and check if there are any clarifying questions that still need to be asked based on the answers you received. Write the final conclusion only if you have checked everything and no more questions remain!!!

Strategy and Work Order

  1. Full Verification
    You must conduct a full, step-by-step analysis of all possible transaction risks, without skipping any stage.

  2. Prohibition on Stopping
    You are prohibited from stopping asking questions and issuing a final conclusion until the following condition is met:
    You have checked ALL key aspects listed below in the "Checklist for Verification"
    It is forbidden to assume or invent information not provided by the user. If information is not provided, consider it a gap and either ask a clarifying question or record it as a risk.

  3. Checklist for Verification
    Before issuing a conclusion, you must mentally check against this list and ensure that for each item either all necessary clarifying questions have been asked and final answers received, or the answer "I don't know" has been recorded as a risk. The list is exhaustive for this transaction:

    ✅ Status and Authority of the Seller: Capacity (individual) / Legitimacy of governing bodies (legal entity). Verification of the document confirming authority (power of attorney, meeting decision, etc.)
    ✅ Basis of the Seller's Ownership Right: Verification of the legality and "cleanliness" of the method by which the seller acquired the property (purchase-sale, inheritance, gift, privatization, annuity, etc.)
    ✅ History of Title Transfers: Verification of the "legal cleanliness" of previous transactions with the object
    ✅ Current Legal Status of the Object: Absence of encumbrances (arrest, mortgage, pledge), claims (court disputes), lease with right of succession
    ✅ Registered Persons: Verification of the presence/absence of persons registered in the property, their legal status, and consent to the transaction
    ✅ Family Status of the Individual Seller: Verification of the regime of joint marital property and obtaining notarized consent of the second spouse (if applicable)
    ✅ Technical Condition of the Object: Compliance with documents (absence of unauthorized renovations)
    ✅ Settlement under the Transaction: A payment mechanism safe for the buyer (e.g., using an escrow account or bank safe deposit box)

  4. Recording "I Don't Know"
    Any "I don't know" answer from the user is automatically recorded by you as a potential risk, which must be reflected in the final conclusion.

  5. Final Check
    Before forming the conclusion, briefly list all the checks performed and ask the user if all necessary aspects are covered and if the user has additional information? After you have asked all questions on all points of the checklist, you must ask the user the last question:
    "That's all my questions. I have checked all points of the checklist. Are you ready to receive the final risk assessment conclusion?"
    Only after the user's affirmative response do you form and output the detailed conclusion.

How AI Helped Create a Landing Page and Legally-Secure Service Logic from Scratch

Experience using DeepSeek for a real task.

I needed to create a landing page from scratch and design the logic of a service in an area with legal restrictions.

  • Write texts for the landing page that explain the essence of the service without violating advertising laws for financial instruments

  • Develop a user flow that excludes signs of licensable activity

  • Ensure built-in mechanisms for compliance with corporate rights

I gave the neural network step-by-step tasks: first - the general structure of the landing page, then - texts for blocks considering legal restrictions. Then - designing the logic of user interaction. After each clarification of legal frameworks (for example, "such and such is prohibited"), the AI restructured the schemes.

Result

I quickly obtained a developed prototype: precise texts for the landing page and detailed user flow diagrams of the registration stages and user journey options for different types of users.

The tool proved effective for rapid prototyping under strict regulatory requirements.

How DeepSeek Helped with Protecting Trade Secrets, Not Just Signing an NDA "for the Tick" 🔒

Sharing a case study on how a neural network helped quickly address a request to review an NDA, turning it from a formality into a working protection tool.

A fintech company asked to "look at" an NDA with a counterparty (a microfinance organization).

At first glance - a standard form.

But during the dialogue, it turned out that the goal was not just to "tick the box," but to actually protect information.

There is no protection if a trade secret regime is not established.

That is, it was necessary to explain to the client that an NDA alone is not enough to hold someone liable for disclosure, and to propose solutions that would strengthen their position.

1️⃣ Risk Analysis in the NDA. I didn't just correct the wording, but with the help of DeepSeek, compiled a table with risks for each contentious point and proposed alternative wordings. For example:

  • Risk: Oral agreements are impossible to prove in court → Solution - introduce a condition for mandatory written confirmation of any orally transmitted information within 3 days.

  • Risk: Inability to recover losses for a leak → Solution - introduce a clear mechanism for paying a penalty.

2️⃣ Instructions on the Trade Secret Regime for the Client: DeepSeek prepared a list of measures that the client needs to implement to be able to hold the receiving party liable for disclosing its trade secrets:

  • 📜 Having a list of information constituting a trade secret and an approved Regulation on Trade Secrets.

  • 👥 Familiarizing employees with this regulation and their NDAs under signature.

  • 🏷 Proper marking of all transferred documents with the stamp "Commercial Secret" with the owner's details.

  • 🔐 Organizational and technical measures: safes, access logs, etc.

The client received not just a corrected document, but a protection strategy: a legally strong NDA + a clear action plan + justification of why it's important.

🎯 Practical Prompt:

If you need to quickly analyze an NDA and give the client real recommendations, not just edits, use this prompt:

You are an experienced lawyer specializing in intellectual property and trade secret protection. Analyze the provided file with the NDA text.

  1. Create a table with risk analysis in the format:

    • Clause Number

    • Risk for the Client

    • Proposed Wording for the Clause

    • Legal Justification (with references to articles of the Civil Code and Federal Law 98-FZ)

  2. In a separate section, prepare recommendations for implementing a trade secret regime:

    • What measures need to be implemented to comply with the trade secret regime (Art. 10, 98-FZ).

    • What additional conditions to include in the NDA (audit, penalty, information destruction).

    • How to document the transfer of confidential information.

Case Study - Contract Analysis in DeepSeek! 🚀

I needed to quickly review a 14-page PDF contract.

I uploaded it directly to DeepSeek in PDF.

I managed not only to quickly conduct an analysis but also to prepare edits that genuinely strengthened the position.

The "Two Prompts" Methodology :)

🔍 STEP 1: AUDIT
First, we ask the AI to identify all risks. Don't ask it to edit, ask it to analyze.

✅ Example prompt:
"You are a lawyer. Check the contract for risks for the Client.
Highlight provisions with a bias in favor of the contractor, explain the risks (financial, legal, operational), indicate clause numbers. Give an overall assessment of the contract."
And I upload the PDF.

The AI will output a structured report.

✍️ STEP 2: EDITING
Based on the analysis, we prepare edits — the neural network suggests wordings for problematic clauses.

✅ Example prompt:
"Based on the analysis, prepare a table with proposed wordings for clauses that minimize risks for the Client.
Give me the wording for each of the contract clauses that eliminates or minimizes risks.
Format - table.
In one column of the table - clause numbers, in the other - the proposed wording for the clause.
For each clause and its proposed wording - a separate row in the table (do not put multiple clauses in one table row)."

The output is a ready-made table for a protocol of disagreements or for inserting new clause wordings into the contract.

Of course, we must not forget to check everything ourselves. But even considering the check, it saves a lot of time!

🤖 How the Neural Network Drew Me a Perfect Project Launch Timeline for a Complex Project in 10 Minutes

Sharing a small case study on how DeepSeek saved me several hours of work and helped beautifully visualize a complex service launch process.

I needed to create a clear launch timeline for an online service for a client.

The process is complex, with many stages, some running in parallel, others having completely unpredictable timelines (e.g., approval by government agencies), and others dependent on the completion of previous ones.

Drawing this manually is time-consuming, I'm unlikely to make it look nice, and it would need to be redrawn every time adjustments are made.

I simply wrote a step-by-step list of all stages to the neural network:

  1. Here are the stages.

  2. Here is their approximate duration.

  3. Here are the critical ones that don't depend on us.

  4. Here are the conditions for the final steps (e.g., payment upon testing the service and receiving approval from the state agency).

The neural network generated a ready-made Gantt chart in a minute.

I made a few adjustments with prompts and got the perfect picture.

The most valuable aspects:

✅ Speed, the whole process - 10 minutes from thought to finished chart.
✅ Clarity. All dependencies, "bottlenecks," and risks are immediately visible. The client understands everything without extra questions.
✅ Flexibility: Did the client ask to swap stages? Add a condition? No need to redraw anything manually. Just tell the neural network: "Move this stage here, and make this one with this condition," and it outputs new code for the chart in seconds.
✅ Focus on risks. The neural network perfectly visualizes stages with unpredictable timelines (highlights them in red), which is critical for lawyers to manage client expectations.

🚀 How AI Found All the Flaws in an Investment Proposal Form in 2 Minutes After a Central Bank Remark

Colleagues who work with the Bank of Russia.

Often, after checking a document, the Central Bank writes a general phrase about non-compliance with the law. For example, in my case: "the form of the investment proposal does not comply with clause 5, article 15 of No. 259-FZ" — and that's it 🤷‍♂️

No explanation of which specific clauses are violated. You have to sit and line-by-line compare the form with the law yourself.

I used DeepSeek-V3 to quickly solve this task.

What the AI did in 2 minutes:

1️⃣ Conducted a line-by-line comparison of the form with each clause of Art. 15 of 259-FZ "On Investment Platforms."
2️⃣ Found non-obvious mistakes that are easy to miss during manual review.
3️⃣ Output a ready-made table with descriptions of shortcomings for each clause and specific edits.

Some of the edits were far-fetched, but many were substantive.

🧠 Neural Network as a Junior Lawyer

A case study on how AI saved me several hours, maybe more.

A work request came in to assess the risks for a licensee acquiring a SaaS product with an option to purchase the code and its deposition with a notary. 🤯

The rightsholder only had a Rospatent software registration certificate, but no agreements with developers, licenses for components, etc. (or they refuse to provide them for verification).

It's clear that the risks are high. But I was asked to formalize a legal opinion, not just outline theses. To describe the risks point by point, give recommendations...
All this is labor-intensive.

I assigned the rough work to DeepSeek.

I made a detailed prompt: described the task, context, and even wrote: "You are an experienced intellectual property lawyer...".

The result impressed me. 🤖

DeepSeek didn't just jot down general phrases, but fully completed the task:

✅ Structured the response into sections: risks for each missing document, risks of the option and code deposition.
✅ Gave due diligence recommendations - a list of documents to request.
✅ Wrote recommendations for the contract regarding guarantees and liability.
✅ Created a quite acceptable and well-formatted document.

My work boiled down to checking it, refining it for the specifics of the deal, and sending it.
Instead of 2-3 hours, I spent 20 minutes.

The neural network is an ideal assistant for routine legal opinions.
It instantly analyzes huge amounts of information and outputs a structured draft.
This is especially valuable for topics with a lot of standard information, but whose collection and compilation take time.

The most important part remains: strategic thinking, verification, client communication, and final decision-making.

🤖 How Neural Networks Help Lawyers in Daily Work: A Personal Case and Prompt!

👋 Sharing an example of how neural networks (in my case, DeepSeek) save time and simplify life for lawyers, especially when friends or relatives ask for help with "non-core" questions.

If you're a lawyer, then for friends and relatives you're also the "go-to lawyer" 😉

Recently, an acquaintance approached me with a problem: she received an order from a management company to restore a ventilation box in her kitchen.

I specialize in financial law, investment platforms, not housing law, but it's hard to say no. 😅

Here's how the neural network helped resolve the issue in 5 minutes:

  1. Uploaded a photo of the order - the system recognized the text and structured the data.

  2. Formulated a prompt - here's an example of the query I sent to the neural network:
    "Recognize the text of the order and prepare a response to it. The response should be written in a formal business style, with a request to provide technical documentation for the ventilation box (parameters, restoration requirements) and to request a commercial proposal from the management company for the work. Add references to the Housing Code and other norms regulating renovations and redevelopments."

  3. Received a ready draft response - all that was left was to check the details and adapt it to the specific situation.

Result: instead of hours of work on the document - a few minutes of preparation, the client (in the person of the acquaintance) is satisfied, and I wasn't distracted from my main work for long.

Such requests are part of the reality for any lawyer: whether you're a specialist in VAT or M&A deals, you'll still be asked questions about utilities, inheritance, or traffic fines.
And here, neural networks are an excellent helper: they don't replace us, but they save time on routine.

🔥 How DeepSeek Helped Form a Table with Investment Platform Reporting for the Bank of Russia

Faced with the task of parsing and sorting a huge list of reporting requirements for the Central Bank. 📊 Instead of manually sifting through documents, I decided to entrust this to a neural network.

I used DeepSeek. 🤖

Here's my recipe for success:

🗣 A clear request is the key to success!

I didn't just write "make a table," but described the structure in detail:

"Make a table where the reporting is sorted by frequency. Columns: 'On a non-regular basis,' 'Quarterly,' 'Annual.' In the rows - the name and number of the form."

✅ Copied the raw text from the Central Bank's instruction into the chat with the neural network
✅ Received a perfect table!

The neural network even correctly handled options with mixed periodicity, placing checkmarks in all the necessary cells. ✔️

A ready-made table in the time it took to copy the text! ⏱️ That's 10 times faster than manual work.

This changes the approach to working with documents. 💪

🚀 How the Neural Network Helped Me Save Time Creating a LegalTech Knowledge Base

I want to share my experience creating a LegalTech knowledge base:

The main problem when collecting information was not a lack of data, but an overabundance.
The internet is flooded with promotional articles, offers, and materials with no specifics or practical value. Manually filtering such content would have been the longest and most tedious stage.

DeepSeek helped me.
It was the primary filter 📝

  1. Intelligent Filtering of Promotional Noise
    I wrote a prompt according to which the AI analyzed the essence of the content, not just paraphrased it.

    "Analyze the provided content (text/article/post).
    Determine if it is promotional (promotes a product/service without objective value and without comparison to others) or useful-educational (contains expert opinions, instructions, case studies, objective reviews).
    If the content is promotional, write its essence in one phrase and mark it as such.
    If it is useful, highlight the key theses and practical insights."

    This approach accelerated content selection and filled the knowledge base only with verified and truly useful information, not hidden ads.

  2. Deep Analysis and Structuring
    For already filtered materials, DeepSeek instantly:

    • Summarized long texts, highlighting the essence.

    • Structured data in a unified format.

    • Helped formulate clear and concise posts for the bot according to a given structure.

Thanks to this symbiosis of human control and artificial intelligence, I managed not just to quickly create a knowledge base, but to fill it with quality content, which is the main value for users.

P.S. If you also struggle with information noise, try this prompt for filtering.
You can evaluate the result and find useful tools for lawyers in the legaltech knowledge base: [link].