🧠 Syngenta AI Agent Hackathon Build

Paradox 2025 — IITM BS

This project is part of a hackathon submission for the Syngenta AI Agent Hackathon held at Paradox 2025, IITM BS.

The goal of Chain_Chan was to design an enterprise-grade knowledge assistant capable of secure, role-aware retrieval and conversation over PDF documents. Traditional RAG systems provide semantic search, but most lack access control at the document or user-role level. This project aimed to solve that gap, particularly relevant for enterprises handling sensitive knowledge bases (legal, pharma, financial, etc).

🔗 Demo Video: View Video
🔗 Problem Statement: View PDF
🔗 Live Demo: soon.

---

🔧 Tech Stack

Layer	Technologies
Gen AI	Python → LangChain, ChromaDB
Backend	FastAPI, MySQL, ChromaDB (Metadata Filtering)
UI	Laravel, PHP, JavaScript, CSS, HTML
Auth/RBAC	Session-based for web app
Chat Storage	MySQL
Architecture	MVC (UI + Laravel backend), emphasis on Controller

---

📐 Design Architecture

"I always wanted to build simple yet elegant systems that do more than they seem to — effortlessly."

• Built with a modular mindset to reuse the AI backend and models in future projects via APIs.
• Inspired by terminal-style UIs with buttons (keyboard navigation planned for future).
• Prioritized functionality over flashy design, resisting the JS/Python complexity trap.
• Chose Laravel + PHP for:
  • Simplicity of integration and library management
  • Strong MVC enforcement
  • Lowest deployment cost
  • Developer familiarity and speed

---

🔗 Data Flow & Connections

There are three tightly coupled chains in the architecture:

1. UI ↔ Traditional Backend (Laravel) → Standard MVC CRUD operations
2. UI ↔ AI Backend (FastAPI + LangChain) → PDF/LLM operations via API
3. AI Backend ↔ Traditional Backend (Laravel) → Uses shared MySQL DB, APIs for auth, sessions

Monolith is possible, but the separation allows API-driven AI components, making them reusable across future projects and platforms.

---

🗂️ Folder Structure & Component Map

graph TD

%% Python Folder
subgraph PY["📁 Python"]
  Classes["📁 classes"]
  App["📄 app.py"]

  %% Classes folder files
  Classes --> super_db["super_db.py"]
  Classes --> super_embed["super_embed.py"]
  Classes --> super_models["super_models.py"]
  Classes --> super_vdb["super_vdb.py"]
  Classes --> super_memory["super_memory.py"]
  Classes --> text_splitting["text_splitting.py"]

  %% super_db.py contents
  super_db --> db_data_init["db_data_init"]
  super_db --> db_query_toolkit["db_quering_agent_toolkit"]

  %% super_embed.py
  super_embed --> embed_model["Init Custom Embedding Model"]

  %% super_models.py
  super_models --> chat_model["Init Custom Chat Model"]

  %% super_vdb.py
  super_vdb --> langchain_vdb["LangChain VDB"]
  super_vdb --> chroma_vdb["Chroma Core VDB"]

  langchain_vdb --> add_embed_pdf["Add + Embed PDF"]
  langchain_vdb --> delete_pdf_embed["Delete PDF + Embed"]
  langchain_vdb --> similarity_search["Similarity Search"]

  chroma_vdb --> role_metadata["Update/Get Role in Metadata"]
  chroma_vdb --> pdf_ops["Delete/View PDF"]

  %% super_memory.py
  super_memory --> retrieve_messages["Retrieve Convo + Messages"]

  %% text_splitting.py
  text_splitting --> text_splitter["Custom Text Splitter"]
  text_splitter --> super_vdb

  %% app.py structure
  App --> FastAPI["FastAPI"]
  App --> pdf_crud["PDF RBAC + CRUD"]
  App --> pdf_llm_rbac["PDF LLM RBAC"]
  App --> llms["LLMs"]
  App --> memory_maintain["Memory Maintain"]
  App --> rbac_rag["RBAC RAG"]
  App --> chat_post["Chat Post"]

  pdf_crud --> super_vdb
  pdf_llm_rbac --> pdf_crud
  pdf_llm_rbac --> laravel_api
  llms --> super_models
  llms --> super_embed
  memory_maintain --> super_memory
  rbac_rag --> super_vdb
  chat_post --> laravel_api
  chat_post --> rbac_rag
  chat_post --> super_db
end

%% Laravel Structure
subgraph LV["📁 Laravel"]
  laravel_api["Laravel API"]
  web_rbac["Web RBAC"]
  chat_crud["Chat CRUD"]

  web_rbac --> laravel_api
  chat_crud --> web_rbac
end

%% MySQL Structure
subgraph DB["🗄️ MySQL"]
  chat_table["Chat Table"]
  session_table["Role/User Session Table"]

  chat_table --> chat_crud
  session_table --> web_rbac
end

Loading

📁 Python Structure

Root Level

• app.py: Entry point of the FastAPI backend, orchestrating various functionalities including:
  • PDF operations
  • Chat interactions
  • RBAC-based LLM access
  • Memory maintenance

---

📁 classes/

🧠 super_db.py

• Handles database interactions for querying and initialization.
• Functions:
  • db_data_init: Initializes data into the DB.
  • db_quering_agent_toolkit: Custom querying toolkit for chat agents.

🔍 super_embed.py

• Manages the embedding model for vector operations.
• Init Custom Embedding Model: Loads and initializes a custom embedding model.

🤖 super_models.py

• Loads the primary conversational model used by LLMs.
• Init Custom Chat Model: Bootstraps a custom LLM for chatting tasks.

🗂️ super_vdb.py

• Abstraction layer over vector databases.

🔢 LangChain VDB:

• Add + Embed PDF: Adds and embeds PDF data.
• Delete PDF + Embed: Deletes embedded PDFs.
• Similarity Search: Performs semantic search across documents.

Chroma Core VDB:

• Update/Get Role in Metadata: Manipulates and retrieves role metadata.
• Delete/View PDF: Manages PDF lifecycle.

🧠 super_memory.py

• Handles conversational memory retrieval and tracking.
• Retrieve Convo + Messages: Manages past chat sessions.

✂️ text_splitting.py

• Manages text chunking.
• Custom Text Splitter: Implements a custom algorithm for text splitting.
• Utilizes super_vdb for downstream vector operations.

---

app.py Functional Blocks

• FastAPI: Web framework that wires all components.
• pdf_crud: PDF-based CRUD operations integrated with RBAC.
• pdf_llm_rbac: LLM-powered PDF operations respecting user roles.
• llms: Interfaces for LLMs using custom models and embeddings.
• memory_maintain: Manages long-term conversation context.
• rbac_rag: Combines RBAC and Retrieval-Augmented Generation (RAG).
• chat_post: Endpoints for interactive chat with LLM and PDF data.

---

📁 Laravel Structure

🌐 Laravel API Layer

• laravel_api: Central REST API layer that interfaces with Python services.
• web_rbac: Implements front-end and middleware-based RBAC logic.
• chat_crud: Handles chat session and message management.

---

##🗄️ MySQL Database Structure

• chat_table: Stores historical chat messages and threads.
• session_table: Maps users and roles to their active sessions.

---

🔄 Data Flow Overview

1. User interacts via Laravel Web Interface.
2. RBAC is enforced via web_rbac.
3. Laravel calls FastAPI (laravel_api) for document processing or chat tasks.
4. FastAPI routes requests to respective handlers in classes/.
5. PDFs are split, embedded, and stored using super_vdb and super_embed.
6. Chat context is managed using super_memory, and DB interactions via super_db.
7. MySQL backs persistent storage for chat and user-session metadata.

---

🔐 Role-Based Access Control (RBAC)

• Enforced in both:
  • Laravel (web_rbac)
  • Python (pdf_llm_rbac, rbac_rag)
• Role metadata is stored in MySQL and embedded alongside documents for context-sensitive access.

---

🤖 LLM + RAG Integration

• Uses custom models for embeddings and chat (super_models, super_embed).
• Retrieval-Augmented Generation (rbac_rag) enhances LLM answers with relevant document context.
• Similarity search ensures high relevance during document querying.