Optimize TheLocalChat for Multilingual PDF AI Processing - 29/10/2024 14:37 EDT - 03/02/2025 11:46 EST
Budget: $250 – $750 USD
Project done in DotNet (no Python, no Java, maybe RestApi to LM Studio).
Optimize a conversational AI model for professional interactions in German and English. Integrate it with "TheLocalChat" for PDF processing. The AI should deliver accurate answers with source citations, merging insights from multiple PDFs for thorough responses. Tasks involve model setup, multilingual handling, enhancing document retrieval accuracy, and ensuring smooth conversation flow. The solution should be efficient on CPU environments, enabling offline functionality.
At the current situation, the model is not acting proper with PDF file content. The PDF to text converter is looking usable. But the interaction of the model to the users command is not usable. This intelligence issue of the model is to get fixed. On the development maschine you have full access, ML Studio is there to test, GIT access to work remote.
Please read the attached ChatGpt Samples.pdf to understand how the model should be able to react. ChatGpt itself is no option because its online and because it can not be trained based on huge PDF collection.
But on our Chat system, we can take time for the PDF data training and also for generating the answer.
There is also a possible idea to just connect to LM-Studio by RestApi, which is running locally and can set custom models.
Optimize a conversational AI model for professional interactions in German and English. Integrate it with "TheLocalChat" for PDF processing. The AI should deliver accurate answers with source citations, merging insights from multiple PDFs for thorough responses. Tasks involve model setup, multilingual handling, enhancing document retrieval accuracy, and ensuring smooth conversation flow. The solution should be efficient on CPU environments, enabling offline functionality.
At the current situation, the model is not acting proper with PDF file content. The PDF to text converter is looking usable. But the interaction of the model to the users command is not usable. This intelligence issue of the model is to get fixed. On the development maschine you have full access, ML Studio is there to test, GIT access to work remote.
Please read the attached ChatGpt Samples.pdf to understand how the model should be able to react. ChatGpt itself is no option because its online and because it can not be trained based on huge PDF collection.
But on our Chat system, we can take time for the PDF data training and also for generating the answer.
There is also a possible idea to just connect to LM-Studio by RestApi, which is running locally and can set custom models.