Changelog
Launch Day with revamped services and UI
Now schedule posts, including images and threads, effortlessly from your browser. Seamlessly integrated with AI content generation tools like Mixtral, GPT-4, and GPT-3.5. Boost your online presence with ease. Stay tuned for more updates!
Adding more value driven AI Agents workflows, multi modal report creation
Building more reliable, value driven AI agents based workflows with a master slave agent architecture.
Building Chrome Extension
Building Chrome Extension to support pdf files at any website on Chrome web browser.
Experimenting with other LLM Models including OS models
Modifying agent orchestration based on experiments with other AI models like claude, gemini, mistral.
Added support for Scanned PDF's and rendering of equations
and formulas in chat box
Added a high-res model for performing OCR on scanned documents.
Added Multimodal Retrieval tool
Using this tool one can now get answers based on text, tabular and figure data simultaneously.
Added button to view file history and improved voice chat latency.
Added share chat option and disabled password
sharing to prevent misuse.
Added ability to provide negative feedback and log results
Added Support for reading and analysis of other document extensions
Supported file extensions have been updated:
.xlxs
.csv
.doc
.docx
.html
.java
.md
.pptx
.rb
.tex
.txt
.css
.js
.sh
.ts
Ran final tests for concurrency, file uploads
Added new pricing based on actual usage
This new pricing is a simple Pay As You Go pricing tier with hosts of benefits.
Running tests for all features, revamped rate limits
Bug Resolution on Frontend
Removed popular bugs on new fe.
New Inference UI published to prod
New Landing Page Design
Brainstormed on new landing pages design.
New UI Design Finalised
Introduced Knowledge Graph creation and triplet based KG creation
Introduced Knowledge Graph functionality:
Utilizes RAPTOR techniques and generating Cypher queries from natural language.
AI workflows II: Create Short form QnA from PDF files.
Instantly generate short form question and answers from each page of your PDF document.
Added AI workflows first one with end to end PDF Document translation
Now translate your PDF documents to 50+ different languages within minutes. Whether it is manuals, SOP's with tables or figures the entire PDF will be translated top to bottom.
Added GPT Researcher and Report Generation from PDF files
With this update, you can now automate Report generation based on your PDF document. Our AI assistant will prepare a comprehensive report for you to review within minutes capturing all information whether tables or text or figures in your PDF files.
Comepletely Revamped Generating Summaries for PDF's
With this update, now get excellent concise summaries of long documents. Our AI assistant uses clustering algorithms to generate holistic summaries that capture the entire context of the PDF document. Better than most on the market as other tools only summarise what can fit within a context window in a single or multiple api call leading to often losing context and accuracy, and missing vital information. Try for free and we guarantee you will be impressed.
Added Web Search Agent
Now one can browse the web while chatting with a document to quickly search for information on the web and view web search results.
Added Code Interpreter Agent
Often while searching for information one requires to perform some calculations. LLM's themselves are not good at performing calculation or logical tasks, therefore a code interpreter agent has been added to write and execute code while performing retrieval. For example, if you were asking our AI assistant to read some values from a specific table and wanted to calculate the mean deviation on those tabular values, then code interpreter Agent would perform this calculation for you.
Added independent image analysis
With this one can chat with images using the Image analysis option from retrieval tools. Just paste the URL of the images and start asking question.
Added Sub-question generator and query augmentation
With this update, you can now ask upto 10 questions all at once and let our AI assistant find answers to all of them in one shot. Complex queries can be broken down to multiple questions which in turn increase contextual relevancy and accuracy.
Recursive Retrieval and building AI Agents flow
Now, you can effortlessly retrieve information from your document using this retrieval method. For example when you need to aggregate results from document, like "what are all the skills mentioned in the CV/Resume". These kind of questions require a more dynamic approach to retrieval and collation of information which is what this feature is for. Started implementing individual AI agents for each retrieval method.
Metadata filtering enabled
Automatically identify metadata filters in chat messages and filter vectorDB based on these. Helps answer question like " explain table with title readings from experiment 1.1".
Added custom metadata generation
Our AI assistant will generate relevant metadata for structuring your pdf content, helps in improved chat experience.
Added ability to chat with figures in documents
Now attach images from top similar images provided at the end of each answer of our AI Assistant. This will help explain figures like bar charts, line charts, diagrams in detail.
Added logic for selecting dyanmic top-k for image and text
With this update, you can now effortlessly plan and automate your social media posts directly from your browser. Schedule content across multiple platforms in advance, ensuring a consistent online presence without the need for manual posting. Our intuitive interface within the Chrome Extension makes it easy to set up and manage your content calendar, all from the convenience of your browser.
First attempt at multimodal system
Image parsing, encoding images, searching within text and image data simultaneously and collating results.
Optimized chunk retrieval with image support
Added support for multiple languages
Added multilingual embeddings for enhanced multilingual support.
Switched VectorDB
Moved to self hosted vector DB to move away from Pinecone as our erstwhile vector database for improved performance and scalability.
Upgraded text embeddings
Upgraded to higher dimension text embeddings.
Implementing Chat Memory
Implemented chat buffer memory using Redis for improved conversation context.
Integrated reranker for more accurate search results
Rerank top-k search results according to user query.
