86 lines
5.4 KiB
Markdown
86 lines
5.4 KiB
Markdown
# LexiChain — Technical Platform Overview
|
|
|
|
## 1. Executive Summary: What is LexiChain?
|
|
**LexiChain** is an advanced intelligence platform specifically designed for the **BFSI** (Banking, Financial Services, and Insurance) sector. It transforms complex, opaque legal documents into interactive, actionable data using a combination of **Generative AI** and **Blockchain Technology**.
|
|
|
|
The core mission of LexiChain is to solve the "Black Box" problem in contracts: where clients and institutions often sign long documents without fully understanding the hidden risks, obligations, or deadlines.
|
|
|
|
---
|
|
|
|
## 2. The Core Problem & Solution
|
|
### The Problem
|
|
* **Cognitive Overload**: Insurance and banking contracts are filled with "Legalese"—dense, technical language that is difficult for non-experts to parse.
|
|
* **Lack of Trust**: There is no easy way to prove that a document hasn't been modified after signing.
|
|
* **Static Data**: Traditional PDFs are "dead" files. You cannot ask a PDF a question like *"What happens if I miss a payment by 3 days?"*
|
|
|
|
### The LexiChain Solution
|
|
LexiChain creates a **"Living Document"** environment. It uses AI to extract meaning and Blockchain to guarantee integrity, allowing users to converse with their contracts in natural language.
|
|
|
|
---
|
|
|
|
## 3. System Architecture
|
|
LexiChain is built using a modern **Distributed Architecture** composed of four primary layers:
|
|
|
|
### A. The Client Layer (Frontend)
|
|
Built with **Next.js 15** and **Tailwind CSS**. It focuses on **User Experience (UX)**, providing a dashboard that works seamlessly on both desktop and mobile. It handles the secure transmission of files to the backend.
|
|
|
|
### B. The Application Layer (Backend)
|
|
This is the "Brain" of the system, powered by **Next.js Server Actions**. It coordinates the flow of data between the user, the database, the AI models, and the blockchain network. It manages authentication, file storage, and the processing pipeline.
|
|
|
|
### C. The Intelligence Layer (AI & RAG)
|
|
This layer uses **Gemini 1.5 Pro** and **Mistral AI** for high-speed analysis. Instead of just "reading" text, it uses a **Vector Database** to perform Retrieval-Augmented Generation (RAG), ensuring the AI answers only based on the specific facts found in the uploaded contract.
|
|
|
|
### D. The Trust Layer (Blockchain)
|
|
A decentralized layer powered by **Ethereum/Hardhat**. It creates a unique cryptographic "fingerprint" (hash) for every contract. Once recorded, this fingerprint becomes an immutable proof of the document's existence and original state.
|
|
|
|
---
|
|
|
|
## 4. How the Application Works (The Pipeline)
|
|
|
|
1. **Intake**: The user uploads a contract (PDF/Image).
|
|
2. **OCR & Parsing**: The system converts the document into machine-readable text.
|
|
3. **Semantic Chunking**: The text is broken down into small "concepts" or chunks.
|
|
4. **AI Analysis**: The AI extracts key metadata (Dates, Parties, Obligations, Risks).
|
|
5. **Blockchain Certification**: The document hash is sent to a Smart Contract to lock in the "Proof of Deposit."
|
|
6. **RAG Indexing**: The chunks are stored in a specialized index for the Chat interface.
|
|
7. **Interaction**: The user can now ask questions, view the blockchain proof, or check their dashboard for upcoming contract deadlines.
|
|
|
|
---
|
|
|
|
## 5. Deep Dive: RAG (Retrieval-Augmented Generation)
|
|
### What is it?
|
|
In simple terms, RAG is like giving the AI a **"Open Book Exam."**
|
|
|
|
Most AI models rely on what they learned during training (which might be old or generic). With RAG, when you ask a question, the system first **searches** your specific contract for the relevant paragraphs, **retrieves** them, and then **gives** them to the AI to summarize.
|
|
|
|
### Why use it in BFSI?
|
|
* **Zero Hallucination**: The AI is forbidden from "guessing." If the answer isn't in your contract, it says "I don't know."
|
|
* **Contextual Accuracy**: It understands the difference between a "Home Loan" in 2010 vs. a "Car Insurance" in 2024 because it only looks at the specific context of your file.
|
|
|
|
---
|
|
|
|
## 6. Deep Dive: Blockchain & Trust
|
|
### The Digital Notary
|
|
In the BFSI world, dates and integrity are everything. If a claim is denied because of a "deadline," the user needs proof that they held the document on time.
|
|
|
|
### How it works technically:
|
|
1. **Hashing**: We turn your PDF into a 64-character string called a "Hash." Even changing a single comma in the PDF would result in a completely different hash.
|
|
2. **Immutability**: Once this hash is written into our **Solidity Smart Contract**, it can never be deleted or changed by anyone—not even the platform administrators.
|
|
3. **Verification**: At any time, a user can "Verify" their document. The system re-hashes the file and compares it to the blockchain. If they match, the document is **Genuine**.
|
|
|
|
---
|
|
|
|
## 7. The Technology Stack (Summary)
|
|
|
|
* **Frontend**: Next.js (React), Tailwind CSS, Lucide Icons, Framer Motion.
|
|
* **Backend**: TypeScript, Prisma ORM, Server Actions.
|
|
* **Database**: PostgreSQL (Neon) for metadata, Vector Storage for AI.
|
|
* **AI**: Google Gemini (Large Language Model), Mistral AI (Fallback with Pixtral Vision).
|
|
* **Blockchain**: Solidity (Smart Contracts), Hardhat (Local Node), Ethers.js (Integration).
|
|
* **Storage**: UploadThing (Secure File Hosting).
|
|
|
|
---
|
|
|
|
## 8. Conclusion
|
|
LexiChain is not just a document viewer; it is a **Decision Support System**. By combining the analytical power of AI with the structural trust of Blockchain, it bridges the gap between complex legal documents and clear, verifiable human understanding.
|