Rabbitt.ai, a leader in AI-powered innovations, is thrilled to announce the launch of ChanceRAG, a cutting-edge no-code Retrieval Augmented Generation (RAG) solution. As Harneet Singh, Chief AI Officer at Rabbitt.ai, announced, ChanceRAG is now positioned as “the enterprise-grade solution for building RAG.”
Designed for simplicity and efficiency, ChanceRAG allows users to upload PDF documents and seamlessly connect their large language models (LLMs) to these documents using a vector database. This innovative solution sets a new standard for retrieval-augmented generation, particularly through its breakthrough Advanced Fusion Retrieval technique, a patent-pending technology by Rabbitt Team and Harneet Singh which merges semantic understanding and keyword matching for superior performance.
Motivation Behind ChanceRAG
Harneet Singh, Chief AI Officer at Rabbitt.ai, commented on the motivation behind this innovation: “LLMs by themselves are not that useful when you are building applications for real-world business use cases like Customer Support Chatbot or an outbound AI Salesperson. You need to build RAG pipelines to get the best out of LLMs, but developing efficient RAG is in itself a challenge. The problem is 2-fold.
At one end, there are a lot of RAG techniques but developers don’t know what works best for their domain and use case. This leads to sub-optimal application performance due to inefficient RAG implementation and a lot of cognitive load on Dev Teams and CTOs.
Secondly, to implement the RAG pipelines for production-grade applications and integrating it with LLMs, existing Tech Stack and Infrastructure requires domain expertise, software engineering knowledge and machine learning and LLM expertise, which is sometimes hard to find in many organisations.
Hence we launched “Chance,” to remove the hit-and-trial and misses out of the RAG pipelines. With Chance, developers and organisations can launch enterprise-grade LLM applications with just a function call. “There is no chance that you can go wrong with our Chance” - said Harneet Singh
We noticed that traditional retrieval methods, whether semantic or keyword-based, weren’t providing the depth and accuracy needed for complex queries. With ChanceRAG, we’ve created a fusion retrieval technique that delivers unparalleled precision and context, something that no current method achieves on its own.”
ChanceRAG addresses the limitations of traditional methods by integrating them for a unified, highly accurate retrieval experience. This innovation is just the start—Rabbitt.ai plans to release additional groundbreaking techniques in the coming weeks, further expanding the capabilities of RAG systems.
Advanced Fusion Retrieval Method: Powering ChanceRAG
The Advanced Fusion Retrieval Method is at the heart of ChanceRAG, our cutting-edge document retrieval system designed to provide accurate, contextually relevant answers. By leveraging a dual-path retrieval system—vector-based and keyword-based methods—it ensures you get the best of both worlds in document search.
How It Works
- User Query and Embedding Creation
It all starts with the user query. Whether you're searching for technical documentation, research papers, or any other type of document, your input is instantly transformed into an embedding vector. This step allows our system to convert your query into a format that can be easily compared against other document vectors.
- Parallel Document Scoring Methods
Once the query embedding is created, our system takes two parallel paths to evaluate the relevance of documents:
- Annoy (Vector-Based Scoring): This method calculates how similar each document is to the query using vector space, enabling it to understand the underlying context and meaning behind the words.
- BM25 (Keyword-Based Scoring): Simultaneously, BM25 applies a traditional keyword search, focusing on term frequency and relevance. This ensures that the documents most relevant to the query's keywords are prioritized.
- Cosine Similarity and Graph Construction
Next, the system builds a cosine similarity matrix, comparing the document vectors to the query. This matrix forms the foundation for a similarity graph, mapping out relationships between documents based on their similarity.
- PageRank for Document Importance
Using this graph, our system applies PageRank, a powerful algorithm that ranks documents by their importance within the graph. This ensures that even if two documents are equally relevant, the one with greater influence or authority is given higher priority.
- Combining Scores for Accuracy
To maximize precision, the scores from the Annoy, BM25, and PageRank processes are combined. By weighting each method (vector-based, keyword-based, and graph-based), our system can effectively balance context, keyword relevance, and document importance.
- Normalization and Ranking
The combined scores are then normalized to ensure consistency across documents, regardless of how they were evaluated. Finally, the system ranks all documents by relevance, ensuring the most valuable results appear at the top.
- Top 5 Results Delivered
The top 5 most relevant documents are returned, providing fast, reliable, and accurate answers to your query, with no room for hallucination or errors.
Benchmarking Excellence
To ensure ChanceRAG delivers unparalleled results, it has been rigorously tested using industry benchmarks:
- Retrieval Evaluation: nDCG@5 = 5, Hit Rate = True, and Precision Rate @K = 80% (avg.).
- Response Evaluation: QA Correctness = True, Hallucinations = False.
These benchmarks confirm ChanceRAG’s capacity for high-precision document retrieval and reliable, accurate responses free from hallucination.
You can try ChanceRAG yourself through a live demo on HuggingFace here: Rabbitt-AI/ChanceRAG.
How ChanceRAG Stands Apart
At the core of ChanceRAG is its proprietary Advanced Fusion Retrieval technique, a patent-pending technology by Rabbitt Team and Harneet Singh, representing a major leap forward in retrieval accuracy and flexibility. By blending vector-based semantic search with keyword-based BM25 methods, this technique ensures users receive the most contextually relevant and precise results possible.
Key Features of ChanceRAG:
- PDF Processing and Text Chunking: Automatically processes PDFs into manageable text chunks for more efficient retrieval.
- Vector Store Creation: Utilizes Mistral embeddings to generate vector representations, stored in a FAISS vector database for fast similarity searches.
- BM25 Index Creation: Complements vector search with BM25 indexing for effective keyword-based retrieval.
- Advanced Fusion Retrieval Function: Combines both methods to deliver highly relevant, accurate responses to user queries.
Interactive Experience with ChanceRAG on HuggingFace
Rabbitt.ai is dedicated to making cutting-edge AI tools accessible to everyone. To demonstrate ChanceRAG’s capabilities, we've created an interactive interface on HuggingFace, using Gradio, that lets users directly experience the power of ChanceRAG in real time.
How to Use:
- Uploading Documents: Begin by uploading your PDF documents. ChanceRAG can handle various types of content, from research papers to technical manuals, providing a robust testing ground for its advanced retrieval features.
- Choosing Response Style: Users can select from multiple response styles—concise, detailed, technical, or creative—depending on their needs. This customizability makes ChanceRAG versatile for different use cases.
- Submitting Queries: Enter your query, and ChanceRAG processes it based on the uploaded document, retrieving not only the answer but also providing insight into the top retrieved documents and their relevance scores.
Upcoming Innovations at Rabbitt.ai
ChanceRAG marks the beginning of a wave of groundbreaking advancements in Retrieval-Augmented Generation (RAG) technology from Rabbitt.ai. In the coming weeks, we will be launching several cutting-edge techniques designed to take RAG systems to the next level:
- Dynamic Query Expansion with Contextual Awareness: This technique intelligently expands user queries by analyzing the surrounding context and related concepts, ensuring that even subtle nuances in the query are captured. It enhances document retrieval accuracy, especially for complex or ambiguous queries, by widening the search without compromising relevance.
- Multimodal Document Summarization: A breakthrough in document summarization, this technique generates concise yet comprehensive summaries from a mix of text, images, and embedded data. This feature is particularly useful for industries dealing with technical manuals, research papers, or complex reports, reducing the cognitive load on users by delivering contextually rich, accurate summaries.
- Adaptive Re-ranking with Real-time Feedback Integration: This dynamic re-ranking system continuously improves the relevance of retrieved documents by learning from user interactions. It adapts in real-time based on user preferences and feedback, providing increasingly refined and accurate results with each query iteration.
- Context-driven Document Segmentation: An advanced segmentation technique that breaks down long documents into coherent, contextually meaningful sections, making it easier for the retrieval system to provide precise answers and insights from highly detailed sources.
These innovations are designed to complement and elevate ChanceRAG, enabling businesses to unlock deeper insights from their data with greater accuracy, speed, and flexibility. Rabbitt.ai remains at the forefront of RAG development, continually pushing the boundaries of what’s possible in AI-driven retrieval solutions.
About Rabbitt.ai
Founded by a serial entrepreneur and IIT Delhi alum, Mr. Harneet Singh, Rabbitt AI is a leading generative AI platform that empowers businesses to build and deploy sophisticated AI applications. With a focus on core generative AI capabilities, custom LLM development, RAG fine-tuning, data-centric AI, MLOps integration, and voice bot AI agents, Rabbitt.ai accelerates AI development, improves model performance, and drives business innovation.