The problem SenseAI solves

In India, the digital divide is a significant issue, particularly in rural areas where around 70% of the population lacks reliable internet access. This divide exacerbates disparities in educational opportunities, with government schools in these regions often unable to match the resources and technological advancements available in private schools. Consequently, students in rural government schools face challenges in accessing quality education, which hinders their overall development and future prospects.

Challenges:

Internet Connectivity: Rural areas suffer from poor internet infrastructure, with many villages having limited or no access to high-speed internet. This limitation makes it difficult for schools to integrate online educational resources and tools effectively.

Electricity and Hardware: Many rural schools face intermittent electricity supply and lack modern hardware, further restricting the use of advanced digital tools.

Disparity Between Government and Private Schools:

Resource Availability: Private schools, often located in urban or semi-urban areas, have better access to technological resources, including internet connectivity, modern devices, and educational software.

Quality of Education: The disparity in resources translates to differences in the quality of education. Private schools can leverage online tools and platforms to enhance learning, whereas government schools are often left behind.

Challenges we ran into

Difficulties:
Finding the offline solution for education while maintaing the efficiency of the product was a difficult job.
Lack of GPU to finetune our own model in our workspace.
More time consumption in creating the dataset from the available resources.
Updation of the information to the model when deployed in a real scenario.
Limited hardware facilities in the school in the rural areas made the task complicated.

Solutions:
Finetuning the model with the constrained resources(book of the students) made to maintain the efficiency of the product.
Implemented several finetuning techniques like PEFT, LORA, Q-LORA in the free source cloud platforms like colab from google.
We have automated the process of dataset creation from different resources using a pipeline which made possible for the users to use their own new updated contents.
Quantizing the model in 4bit solved the problem of resourse usage which made us to run the model in a cpu environment with a processor equal to intel i5 processor and a RAM of 8 GB.