POPULAR FEATURES
Unified Multi-Model Platform:
A single platform that combines OpenAI, Claude, Gemini, and LLaMA models, with subscription options for both individuals and teams.
Production-level RAG:
Accept a variety of file types and support advanced document chat to enable multiple files in one single chat.
Chrome Extension:
An intuitive interface that integrates smoothly with Chrome to summarize YouTube videos, generate transcripts and provide browser-based features.
Deftgpt Spaces:
A space for teams to collaborate with unlimited members, streamlined billing, and improved productivity.
PROJECT REQUIREMENT
DeftGPT developed a comprehensive AI platform with advanced features specifically tailored for users in China.
- The platform needed to integrate multiple models like OpenAI, Claude, Gemini, and LLaMA for a unified experience.
- A robust RAG system was required to handle high traffic and support a variety of file types.
- Seamless interaction with images, videos, and audio files was essential for advanced media understanding.
- A Chrome extension was included to summarize YouTube videos, generate transcripts, and provide additional browser-based features.
OUR APPROACH
We took a strategic approach, using modern technologies and efficient processes to meet the project’s needs.
Backend & Database
We used FastAPI for the backend to ensure high performance in real-time data handling and seamless scalability.
Model Diversity
Integrated advanced models such as OpenAI, Claude, Gemini, and LLaMA into a single unified interface.
Scanned PDF
Implemented advanced document chat with support for scanned PDFs by converting them to images and processing them with GPT vision models.
Data and Media Processing
Built advanced tools to analyze CSV and XLSX files while seamlessly handling images, videos, and audio for enhanced interaction.
Chrome extension
Implemented features for summarizing YouTube videos, generating transcripts, and more to enhance user experience directly in the browser.
PROJECT SUCCESS
The following were successfully delivered:
- Leading AI models were prioritized through extensive R&D to meet the quality and versatility requirements, ensuring optimal performance.
- The RAG pipeline was optimized by shifting from ChromaDB to MilvusDB, effectively handling high system traffic and improving efficiency.
- Smooth platform performance was ensured by processing all functionalities asynchronously, providing a seamless experience for large-scale applications.
- Extensive R&D on open-source models for video understanding led to the selection of Gemini1.5 for its superior performance, which was then integrated into our RAG pipeline.
- Real-time performance challenges were tackled by optimizing RAG workflows and utilizing efficient storage and retrieval systems like MilvusDB to handle large files and media.
- The challenges of processing scanned PDFs were addressed by optimizing the pipeline with pdftoimage conversion and vision models for efficient analysis.
RESULTS
DeftGPT successfully launched as a powerful AI platform designed specifically for users in China.
It offered robust RAG capabilities, advanced media understanding, and seamless browser integration.
As OCloud Solutions' first AI project, it included more integrations with LLMs than any other project. This set a new standard for AI workflows and highlighted our expertise in delivering innovative solutions.
In addition to addressing local needs, the platform handled Chinese-language data efficiently through its advanced RAG pipeline. It paved the way for future projects with cutting-edge features and enhanced AI capabilities.
TECHNOLOGIES WE USE
React
Python
FastAPI
PHP
Docker
AWS
MilvusDB
MySQL
Langchain
AWS S3
Open-AI
anthropic
replicate
Stability.ai
Gemini
Search API