The problem ViduraAi solves

It solves the problem of Lack of Legal Awareness and legal knowledge amongst the common people and especially people belonging from marginalized communities who have limited resouces and educating them in a gamified way by using story videos created using Ai tools like invideo, and then we'll ask questions based regarding the story of that particular right or law presented in the Video. Adding a streak system and awarding people with different badges depending on their streaks and lessons they've taken similar to platforms like typing.com
For Answering Legal Queries - People can come to our web platform and ask for their questions and discuss their situations and our ml model (Fine-Tuned BERT) will answer to their queries,
We've prepared our custom dataset for BERTQuestionAnswering lib, containing question and answers from a given context taken from Constitution of Indian.pdf

Challenges we ran into

Fine Tuning our BERT model (bert-base-uncased) using simpletransformers library which based on transformers library of Hugging Face , which is a LLM and can be accessed through transformers library and we can fine tune it on our own dataset. The problem we faced was an error while training our model , our train.json and test.json dataset were raising issues of context not found instead we provided clear context for them, and this breaks our program from further execution.
The possible solutions we've tried were:
We've seen various youtube tutorials, official documentation of simpletransformers and used chatgpt to debug our code, but it says that our code is right still facing that pariticular issue.