https://github.com/mohith202/ema-chatbot
A public chatbot can use PDF from user or uses preloaded dataset and answer Query from user while displaying source.
https://github.com/mohith202/ema-chatbot
embeddings-word2vec lama llm pdf-document-processor python3 voice-to-text
Last synced: 7 months ago
JSON representation
A public chatbot can use PDF from user or uses preloaded dataset and answer Query from user while displaying source.
- Host: GitHub
- URL: https://github.com/mohith202/ema-chatbot
- Owner: Mohith202
- Created: 2024-06-17T14:43:46.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2025-03-22T06:55:48.000Z (7 months ago)
- Last Synced: 2025-03-22T07:27:14.643Z (7 months ago)
- Topics: embeddings-word2vec, lama, llm, pdf-document-processor, python3, voice-to-text
- Language: Python
- Homepage: https://ema-model-89t7szg5bruabesgecwsqu.streamlit.app/
- Size: 41 MB
- Stars: 0
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
**PROCESS PLANNING OF CHATBOT**
**More detail of the process can be found in "Process_planning_of_chatbot.docx: https://docs.google.com/document/d/1TT4Uw04-QSFhzyTF74MQCTh00UnZ14x_/edit?usp=sharing&ouid=106890396692261549451&rtpof=true&sd=true " and live project link is :https://ema-chatbot.streamlit.app/**
**Step-by-Step Process**
**1. Requirements:**
• Data collection: gone through many articles to finalizing a best article and converted to pdf using AI tool.
• API selection: search for various model like GPT-3 by openAI ,transformer from hugging face and langchain. Finalized langchain for its modular and flexible framework for building advanced NLP applications with a focus on chaining operations and managing conversational context.
• Model selection: Together offers many model in which meta-llama/Llama-3-70b-chat-hf is collection of pretrained and instruction tuned generative text models in 70B sizes from hugging face. The cost is also average ($0.9) per 1M token.
• Finalized LangChain for document loading, text splitting, and embeddings.
**2. Data Preparation**
Data preparation involves collected PDF, loading, and processing the documents to make them suitable for query-answering tasks.
Loading PDF Documents:
• Used PyPDFLoader from langchain_community.document_loaders to load documents from both files and folders.
**3. Text Splitting**
Splitting Documents:
• Used RecursiveCharacterTextSplitter to split documents into manageable chunks (2242) for processing.
**4. Embeddings and Vector Store**
Generating Embeddings:
• Used SentenceTransformerEmbeddings to convert document chunks into embeddings.
• An embedding function was used to converts text data into numerical vectors (embeddings) that can capture the semantic meaning of the text. So that it make easy to integrate with vector stores.
Vector Store:
• Used Chromadb for storing and retrieving document embeddings.
**5. Language Model Integration**
Using Together API:
• Together LLM was used for generating responses based on queries.
**6. Building the Query Agent**
Prompt Template:
• Defined a prompt template for the LLM to generate responses.
RetrievalQA Chain:
• Used the RetrievalQA chain from LangChain to create a query-answering pipeline.
Conversational Agent:
• Implemented a class to handle the conversational context and history.
**7. Streamlit User Interface**
Building the UI:
• Used Streamlit to create a simple user interface for uploading files and entering queries so that easy testing/query input can be done.
**Future Enhancements:**
• Can add a list for a topic related search. So that only relevant topics are used.
• Optimize the data preparation process so that a faster query can be generated.
• Can add logs so that if its crash, it should be easy to debug.
**Language Used:**
