Spaces:
Running
Running
Update about.md
Browse files
about.md
CHANGED
|
@@ -3,23 +3,23 @@ Built with Langchain, Hugging Face, and RAG-powered LLMs
|
|
| 3 |
|
| 4 |
I developed a robust, AI-driven Q&A application that reads and comprehends financial reports in PDF format to deliver insightful, accurate answers along with reasoning. This project showcases the fusion of document intelligence, vector search, and large language models to simplify financial data exploration and analysis.
|
| 5 |
|
| 6 |
-
## ✅ Project Overview:
|
| 7 |
The application is built to process and answer questions from quarterly financial reports of Titan Company Ltd and Hindalco Industries Ltd, covering a period from Q1 FY2024 to Q3 FY2025. These documents include a mix of narrative audit reports and structured financial tables, such as balance sheets and profit & loss statements.
|
| 8 |
|
| 9 |
-
## 🔍 Core Workflow and Architecture:
|
| 10 |
Document Ingestion and Processing
|
| 11 |
Leveraged Langchain's UnstructuredPDFLoader to ingest PDF documents. This component efficiently extracts both textual narratives and tabular data, enabling a unified representation of structured and unstructured financial content.
|
| 12 |
|
| 13 |
-
## Semantic Embeddings & Vectorization
|
| 14 |
Transformed the extracted data into high-dimensional embeddings using two top-tier Hugging Face models:
|
| 15 |
sentence-transformers/gtr-t5-large
|
| 16 |
BAAI/bge-large-en-v1.5
|
| 17 |
These embeddings were stored and indexed using FAISS, enabling fast and accurate similarity-based retrieval.
|
| 18 |
|
| 19 |
-
## Retrieval-Augmented Generation (RAG) Chatbot
|
| 20 |
Built an intelligent chatbot using Langchain’s HuggingFace RAG pipeline, powered by the cutting-edge mistralai/Mixtral-8x7B-Instruct-v0.1 model. This allows the chatbot to fetch relevant document fragments and generate contextual, reasoned responses to user queries.
|
| 21 |
|
| 22 |
-
## Deployment
|
| 23 |
The entire application was deployed seamlessly on a Hugging Face Space, offering an intuitive chat interface for users to explore financial insights in real-time.
|
| 24 |
|
| 25 |
# 💡 Key Features:
|
|
|
|
| 3 |
|
| 4 |
I developed a robust, AI-driven Q&A application that reads and comprehends financial reports in PDF format to deliver insightful, accurate answers along with reasoning. This project showcases the fusion of document intelligence, vector search, and large language models to simplify financial data exploration and analysis.
|
| 5 |
|
| 6 |
+
## 1. ✅ Project Overview:
|
| 7 |
The application is built to process and answer questions from quarterly financial reports of Titan Company Ltd and Hindalco Industries Ltd, covering a period from Q1 FY2024 to Q3 FY2025. These documents include a mix of narrative audit reports and structured financial tables, such as balance sheets and profit & loss statements.
|
| 8 |
|
| 9 |
+
## 2. 🔍 Core Workflow and Architecture:
|
| 10 |
Document Ingestion and Processing
|
| 11 |
Leveraged Langchain's UnstructuredPDFLoader to ingest PDF documents. This component efficiently extracts both textual narratives and tabular data, enabling a unified representation of structured and unstructured financial content.
|
| 12 |
|
| 13 |
+
## 3. Semantic Embeddings & Vectorization
|
| 14 |
Transformed the extracted data into high-dimensional embeddings using two top-tier Hugging Face models:
|
| 15 |
sentence-transformers/gtr-t5-large
|
| 16 |
BAAI/bge-large-en-v1.5
|
| 17 |
These embeddings were stored and indexed using FAISS, enabling fast and accurate similarity-based retrieval.
|
| 18 |
|
| 19 |
+
## 4. Retrieval-Augmented Generation (RAG) Chatbot
|
| 20 |
Built an intelligent chatbot using Langchain’s HuggingFace RAG pipeline, powered by the cutting-edge mistralai/Mixtral-8x7B-Instruct-v0.1 model. This allows the chatbot to fetch relevant document fragments and generate contextual, reasoned responses to user queries.
|
| 21 |
|
| 22 |
+
## 5. Deployment
|
| 23 |
The entire application was deployed seamlessly on a Hugging Face Space, offering an intuitive chat interface for users to explore financial insights in real-time.
|
| 24 |
|
| 25 |
# 💡 Key Features:
|