Indic focused SLM/LLMs

Is anyone here who has built Indic-focused SLM/LLM models for text generation? If yes, please DM me. Want to understand the following things

1. Want to understand the training/fine-tuning and running costs.
2. I also want to understand how much training data was required to attain minimum acceptance criteria.
3. Was the effort worth the reward? Meaning that had you not built the model yourself and used any other foundational models, how could the cost and output differed?