Kalami (beta): Half-Million Token Size in Your ChatGPT Window ๐
๐ข ChatGPT Window Integration: Enhances your learning and research experience by enabling the analysis of over 100 pages of PDFs or multiple file formats, seamlessly within your ChatGPT window.
๐ข Multiple Data Sets: With Kalami's 500,000 token capacity, you can search, analyze, and index multiple files at once, boosting your learning and debugging abilities.
๐ข File Upload Flexibility: Choose from three convenient file upload options: PDF, paste a URL, or directly copy and paste the data set.
๐ซ No APIs, no code required.
Think of Kalami as the cargo ship ferrying vast amounts of data from its source to the island of ChatGPT ๐๏ธ
Fantastic Use Cases โจ
Developers ๐ป
โ Seamlessly switch between large code repositories within your ChatGPT window ๐ก
- Code Repository: Upload entire GitHub repositories with ease thanks to the 20x token size enhancement.
- Debugging Aid: Reference extensive code snippets and trace bugs effectively.
- Multi-contextual Queries: Invoke multiple contexts and reference them in a single query, streamlining the debugging process.
Students ๐
โ Transform ChatGPT into your customized study companion ๐ก
- Study Partner: Upload textbooks or lecture notes to assist with homework, coursework or test preparation.
- Multi-Source Learning: Cross-reference numerous textbooks, articles, or online resources for a comprehensive understanding.
- Interactive Learning: Engage in interactive study sessions, probing uploaded materials with various queries.
Installing & Using Kalami โ
Download & Install:
- Download the build file.
- Go to chrome://extensions/, click load unpacked and select the downloaded file.
Activate & Upload:
- Open a ChatGPT window and click the new Kalami button on the right.
- Choose 'Upload Context'. upload your file, name it, and enable.
Query:
- Type a context-related query and click "Ask Kalami" found in the bottom right.
Welcome to Kalami.ai
Inspiration ๐
โถ Kalami began as a concept for a chat-sharing app, a platform that let users share and extend conversations. Initially, we hit obstacles as there was no straightforward way to connect to someone else's conversation. But, we had an epiphany - what if we used someone's existing chat as the context for a new conversation?
โถ We set out to develop a system capable of scraping, vectorizing, and organizing data using Pinecone and Langchain. The goal was to make this context retrievable and add depth to new chats. This methodology showed promise, particularly for users who had trained a model on a specific task, like content writing.
โถ During the development phase, we employed the app in our coding process. The outcome? We could transfer vast amounts of data to a chat with GPT-4 and were freed from the tedious task of copying and pasting new information. This led us to wonder - what if we could add more than just text?
โญ Hence, we pivoted. The redefined objective of Kalami was to simplify the process of arming GPT-4 with information it hadn't encountered before.
What it does ๐จ
Here is an illustrative diagram to outline the elegance of Kalami's simplicity...
This workflow ensures that information is readily available for reference, creating a seamless and insightful conversation.
Limitless Applications ๐
Researchers ๐งช
โ Navigate through vast literature and cross-reference effortlessly ๐ก
- Research Paper Repository: Load multiple academic papers into ChatGPT for swift cross-referencing and comparison.
- Consolidated Knowledge: Combine knowledge from multiple papers or datasets to draw connections and make new inferences.
- Dedicated Study: Upload textbooks and dissect complicated theories with the assistance of AI, at your own pace.
Project Managers ๐ ๏ธ
โ Centralize project documents and derive insights efficiently ๐ก
- Task Management: Upload project details and use ChatGPT to keep track of tasks, deadlines, and responsibilities.
- Meeting Minutes: Analyze meeting transcripts and summarize them into actionable tasks or highlight key decisions.
- Cross-Reference Projects: Compare current projects with past project documents to learn from past experiences or mistakes.
Content Creators ๐ฐ
โ Spark creativity by cross-referencing a multitude of content sources ๐ก
- Content Ideation: Cross-reference multiple pieces of content for inspiration or ideas for your next project.
- Competitor Analysis: Compare your content against top-performing content in your niche.
- Creative Support: Use ChatGPT to brainstorm or polish your content using the loaded materials.
Entrepreneurs ๐ก
โ Dive deep into patent databases and fuel your invention process ๐ก
- Patent Research: Upload and search through multiple patents swiftly for novelty checking.
- Invention Brainstorming: Load scientific papers or industry trends for idea generation and validation.
- Technical Guidance: Refer to related patents or technical papers while designing your invention.
Financial Analysts ๐
โ Make informed financial decisions with instant access to vast financial data ๐ก
- Financial Data Analysis: Upload and reference financial reports, balance sheets, or industry reports for thorough analysis.
- Market Trend Tracking: Keep up with financial trends by loading and querying various market research reports.
- Investment Decisions: Compare multiple company financials for investment decision-making.
How we built it ๐ง
Our stack:
- OpenAI's GPT-4 (impressive model size of 175 billion tokens).
- Langchain (used to parse, vectorize, and scrape a webpage).
- Pinecone (a vector search engine, used to retrieve the vectorized data).
Challenges we ran into ๐ง
Kalami wasn't without its challenges:
- Resource Intensity: The process of vectorizing large amounts of data can be resource-intensive.
- Code Limitations: Navigating around the limitations of GPT-4 when handling large codebases.
Despite these challenges, our perseverance paid off.
Accomplishments that we're proud of ๐๏ธ
We are proud to have transformed Kalami from a chat-sharing app concept into a robust tool that helps users upload large files and arm GPT-4 with information it hadn't encountered before.
What we learned ๐
The journey with Kalami has broadened our understanding of AI and its potential, from merely sharing chat data to delivering customized, insightful experiences.
What's next for Kalami? ๐ญ
Building crazy features ๐
๐ซ GitHub Integration: We are enhancing Kalami with the ability to reference GitHub repositories directly in your ChatGPT window. This feature will allow you to draw from your codebase during your AI chats.
๐ซ API Modules: Our team is developing a feature to reference API requests within ChatGPT, opening new possibilities for dynamic and interactive AI chats.
๐ข Expanding Data Types: We aim to support more data types in the near future. This includes various formats of media and document files for even richer and more context-aware conversations.
๐ข Improved Token Management: To further optimize your interaction with the increased token capacity, we are designing more advanced token management strategies.
๐ข User Interface Enhancements: In pursuit of the perfect user experience, we are continually refining and updating Kalami's user interface. You can look forward to a more intuitive and efficient chat experience.
Built With
- chatgpt
- google-cloud
- javascript
- langchain
- mongodb
- nextjs
- node.js
- pinecone
- prompt-engineering
- python
- vercel
- web-scraping
Log in or sign up for Devpost to join the conversation.