🤖 MinerlexAI 3.0

Revolutionizing Mining Law with AI-Powered Precision
🏛️ 🇮🇳 Multilingual Legal Assistant | 📚 RAG + Reranking | 🎙️ Voice + Translation | 💡 Gemini + Ollama + FAISS

---

🧠 What is MinerlexAI?

MinerlexAI 3.0 is your intelligent legal research assistant tailored for Indian mining laws. It reads your PDFs, understands your questions, and delivers clear answers in your chosen Indian language—with voice support.

🧾 Ideal for legal researchers, activists, tribal communities, government officers, and NGOs working in the mining and legal ecosystem.

---

✨ Key Features

🔍 Document Understanding
‣ Upload mining law PDFs and auto-index with semantic chunking
‣ Uses PDFPlumber + FAISS for RAG-based retrieval

🧠 Smart AI Answers
‣ Uses Gemini + DeepSeek for precise legal Q&A
‣ 3x candidate generation + semantic reranking using MiniLM-L6-v2

🌐 Multilingual & Inclusive
‣ Translate answers into 12+ Indian languages via MBart50
‣ Supports English, Hindi, Bengali, Tamil, Telugu, Marathi, and more

🔊 Text-to-Speech
‣ Converts both English & regional text answers to voice
‣ Empowers low-literacy communities with spoken legal access

🎙️ Speech Input
‣ Speak your question instead of typing
‣ Uses SpeechRecognition and Google ASR

---

🖼️ UI Preview

---

⚙️ Tech Stack

Area	Tools & Models
Frontend UI	Streamlit (custom dark theme)
LLMs	Gemini 1.5 Flash, DeepSeek via Ollama
RAG	LangChain + FAISS + PDFPlumber
Translation	MBart50 Many-to-Many (HuggingFace)
Embeddings	Ollama (DeepSeek R1)
Answer Reranking	SentenceTransformers (MiniLM-L6-v2)
Voice Synthesis	gTTS (Google Text-to-Speech)
Speech Input	SpeechRecognition + Google API

---

📁 Folder Structure

minerlexai/
├── main.py                # Streamlit app
├── .env                   # Your API key for Gemini
├── requirements.txt       # All Python dependencies
└── README.md              # You're reading it!

---

🔧 Getting Started

1️⃣ Clone the repository

git clone https://github.com/yourusername/minerlexai.git
cd minerlexai

2️⃣ Create a virtual environment

python -m venv venv
source venv/bin/activate   # Windows: venv\Scripts\activate

3️⃣ Install dependencies

pip install -r requirements.txt

4️⃣ Add your Gemini API key to .env

GEMINI_API_KEY=your_api_key_here

5️⃣ Login to Hugging Face

huggingface-cli login

6️⃣ Run the app

streamlit run main.py

---

🌍 Supported Languages

Language	Translation ✅	Audio 🔊
English	✅	✅
Hindi (हिन्दी)	✅	✅
Bengali (বাংলা)	✅	✅
Tamil (தமிழ்)	✅	✅
Telugu (తెలుగు)	✅	✅
Marathi (मराठी)	✅	✅
Gujarati (ગુજરાતી)	✅	✅
Malayalam (മലയാളം)	✅	✅
Kannada (ಕನ್ನಡ)	✅	✅
Urdu (اردو)	✅	✅
Odia (ଓଡ଼ିଆ)	✅	❌
Assamese (অসমীয়া)	✅	❌

---

🎨 MinerlexAI 3.0 Sequence Diagram

sequenceDiagram
    participant User
    participant UI as Streamlit UI
    participant PDF as PDF Processor
    participant FAISS as Vector Store
    participant RAG as Ollama RAG
    participant Gemini as Gemini API
    participant Rerank as SentenceTransformer
    participant MBart as MBart Translator
    participant gTTS as gTTS Audio
    participant Output as Output Display

    User->>UI: Open App
    User->>UI: Upload PDF
    UI->>PDF: Extract and Split Text
    PDF->>FAISS: Create Vector Store

    User->>UI: Ask Question / Speak
    UI->>FAISS: Check if PDF is Indexed
    alt Context Available
        UI->>FAISS: Retrieve Relevant Chunks
        UI->>RAG: Generate Multiple Answers
        RAG->>Rerank: Rerank with MiniLM
        Rerank->>UI: Return Best Answer
    else No Context
        UI->>Gemini: Query via Gemini API
        Gemini->>UI: Return Answer
    end

    UI->>MBart: Translate Answer
    MBart->>gTTS: Generate Audio (EN & Local)
    gTTS->>Output: Return Audio Files
    UI->>Output: Display Answer + Audio

Loading

---

🎨 MinerlexAI 3.0 Flowchart Diagram

flowchart TD

    %% === User Interaction ===
    UI["Streamlit Web App"]
    UI --> Action{{User Action}}

    %% === Document Upload ===
    Action -->|Upload PDF| Upload[Upload Mining Law PDF]
    Upload --> Extract[Extract Text from PDF]
    Extract --> Chunk[Split into Chunks]
    Chunk --> Embed[Create Embeddings]
    Embed --> Vector[Store in Vector Database]

    %% === Question Input ===
    Action -->|Ask Question or Use Mic| Query[User Input]
    Query --> ContextCheck{{Is PDF Indexed}}

    %% === With Context ===
    ContextCheck -->|Yes| Retrieve[Search Relevant Chunks]
    Retrieve --> FormatPrompt[Prepare Context Prompt]
    FormatPrompt --> Generate[Generate Answers]
    Generate --> Rerank[Rerank Answers]
    Rerank --> BestAnswer[Select Best Answer]

    %% === Without Context ===
    ContextCheck -->|No| Gemini[Generate Answer from Gemini]
    Gemini --> BestAnswer

    %% === Translation and Audio ===
    BestAnswer --> Translate[Translate Answer]
    Translate --> AudioEN[Generate English Audio]
    Translate --> AudioLocal[Generate Translated Audio]

    %% === Final Output ===
    AudioEN --> Output[Show Answer and Audio]
    AudioLocal --> Output

    %% === Optional Mic Input ===
    Query -->|Mic Input| Speech[Transcribe Speech to Text]
    Speech --> Query

    %% === Visual Groups ===
    subgraph Document Handling
        Upload --> Extract --> Chunk --> Embed --> Vector
    end

    subgraph Question Answering
        Retrieve --> FormatPrompt --> Generate --> Rerank --> BestAnswer
        Gemini --> BestAnswer
    end

    subgraph Translation and Audio
        BestAnswer --> Translate --> AudioEN
        Translate --> AudioLocal
    end

Loading

---

🔍 Key Components in Flow:

Component	Role
Streamlit	Frontend for UI and interaction
PDFPlumber	Extracts text from PDF
RecursiveCharacterTextSplitter	Splits text into semantically meaningful chunks
FAISS	Stores and retrieves embeddings
Ollama + DeepSeek	Generates context-aware answers
SentenceTransformers	Reranks answers semantically
Gemini	Fallback LLM when no document is uploaded
MBart50	Translates output into user-selected Indian language
gTTS	Converts translated or original answer into speech
SpeechRecognition	Captures and transcribes user's spoken query

---

🧠 How the RAG + Reranking Works:

1. Retrieve: Use FAISS to fetch top-k relevant chunks from PDF based on query.

2. Generate: Pass each to the Ollama-powered RAG prompt template.

3. Rerank: Compare similarity of each generated answer to the original query using cosine similarity → choose the highest.

---

🎯 Use Cases

• ⚖️ Legal research and interpretation
• 🧑🏽‍🌾 Grassroots education and tribal rights
• 🏛️ Policy making and compliance for mining law
• 🧩 NGO support for land & environment advocacy
• 🗣️ Inclusive access to law via voice

---

📌 What's Next?

• Multilingual support + TTS
• RAG + Answer reranking
• Web deployment (Streamlit Cloud / Vercel)
• Fine-tuning on Indian legal data
• Visual QA from scanned government docs
• Build-in chatbot & memory

---

📜 License

This project is licensed under the MIT License.

---

🧠 Acknowledgements

• Google Gemini
• Ollama
• Hugging Face Transformers
• LangChain
• Streamlit

---

🤖 MinerlexAI
Built with 🇮🇳 vision, powered by AI, for a future that's smarter, fairer, and just.