All countries

Price: 0

Number of applications: 3

Decision acceptance deadline

16.03.26 (inclusive)

Form of award

By agreement

Product status

MVP

Task type

ICT tasks

Сфера применения

Media sphere

Область задачи

Intelligent control systems

Type of product

Mobile app

Problem description

Conditions: Budget: up to 120,000 rubles (payment is made in stages over the weeks). Duration: 4 weeks. Work through my GitHub (daily commits). The main criterion: Experience working with RAG and long legal texts.

Expected effect

, FINAL TOR FOR 1 MONTH (For the contract) Sprint 1: Data entry and OCR (7 days) Functionality: Telegram bot accepts PDF (text or scanned). Technology: Using Marker or PyMuPDF + Tesseract/EasyOCR to clear text from "junk". Result: The bot outputs structured JSON: the case number, the court, the parties, and the operative part. Sprint 2: Legal Intelligence / RAG (7 days) Functionality: Search for "leads". The model compares the solution with the articles of the CPC/APC and the Plenum database. Technology: Vector search in the database of scientific information. Industrial engineering: "Find 3 violations of procedural law." Result: A message in the bot: "Chance of cancellation: 75%. An error was found in art. 330 of the CPC." Sprint 3: Document generation (7 days) Functionality: Assembling an appeal using a template. Technology: python-docx library. The use of masks (Plaintiff, Defendant, Arguments). Result: The "Download .docx" button. The file is ready for minimal editing by a lawyer. Sprint 4: Debugging and Prediction (7 days) Functionality: Setting the final accuracy. Assembling error logs. Result: The system has been tested on 20 real cases. The accuracy of the argument has been confirmed by a lawyer. Preparation of a demo video for the investor.

Full name of responsible person

Ilshat

Purpose and description of task (project)

Create a core system for analyzing court decisions and generating appeals in 30 days. Stack: Python, LangChain/LlamaIndex, Vector DB (Chroma/Qdrant), YandexGPT/GPT-4o API. What needs to be done (MVP): OCR module: High-quality parsing of PDF scans (court decisions). RAG-pipeline: Indexing the database of laws and Plenums of the Armed Forces for accurate references. Reasoning: Analyzing the referee's mistakes and calculating the odds (Predict). Docx generator: Output of the finished appeal in Word.

Note

Send examples of your AI bots/projects with documents to BOS. @Fghhhh556