millat/indian_university_guidance_for_bangladeshi_students

Indian University Guidance for Bangladeshi Students Dataset

Dataset Description

This dataset contains 7,044 high-quality, instruction-formatted Question-Answer pairs designed for fine-tuning Large Language Models (LLMs). The primary goal of this dataset is to create a specialized AI counselor that provides accurate, culturally relevant, and comprehensive guidance on Indian universities for Bangladeshi students.

The dataset was generated through a sophisticated, multi-stage pipeline known as SetForge, which transforms unstructured web data into a clean, structured, and context-aware format suitable for generating high-fidelity training data.

• Source Repository: codermillat/SetForge
• Data Collection Tool: codermillat/WebScrape

Dataset Structure

The dataset is provided in a .jsonl (JSON Lines) format, where each line is a distinct JSON object representing a single Q&A pair.

Data Fields

Each JSON object has the following structure:

• question: (string) The question posed, typically from the perspective of a Bangladeshi student.
• answer: (string) A comprehensive, direct answer to the question.
• context: (string) A brief description of the context or topic the Q&A pair belongs to.
• source: (string) The source of the information used to generate the answer.
• metadata: (object) A nested object containing additional metadata for advanced filtering or analysis:
  • degree_equivalence: (string) Information on degree equivalencies, if applicable.
  • grading_conversion: (string) Information on converting grading systems (e.g., GPA to percentage), if applicable.
  • country_origin: (string) The country of origin for the student persona (e.g., "Bangladesh").
  • tone: (string) The intended tone of the answer (e.g., "informative").
  • cultural_sensitivity: (boolean) A flag indicating if the Q&A pair involves cultural-specific context.

Example Record

{
  "question": "I have a GPA of 3.8 and want to study BBA at Sharda. Will I get a 50% scholarship or a 20% scholarship?",
  "answer": "You will receive a 50% scholarship on your tuition fees. Since your GPA of 3.8 out of 5 (≈ 76% ≈ CGPA 7.6/10 in India) is above 3.5, you are eligible for the 50% scholarship. The BBA program is listed under the programs that qualify for the 50% scholarship for Bangladeshi students at Sharda University. Keep in mind, you need to maintain 75% attendance and pass your exams each year to continue receiving it.",
  "context": "Sharda University Scholarship policy for Bangladeshi students and BBA program eligibility.",
  "source": "Sharda University Scholarship Policy",
  "metadata": {
    "degree_equivalence": "N/A",
    "grading_conversion": "GPA 3.8/5 ≈ 76% ≈ CGPA 7.6/10",
    "country_origin": "Bangladesh",
    "tone": "informative",
    "cultural_sensitivity": true
  }
}

Data Curation

The creation of this dataset was a rigorous, research-oriented process designed to maximize quality and relevance. The process was orchestrated by the SetForge pipeline, which is divided into three main parts.

Part 1: The Knowledge Forge (Data Structuring)

1. Data Collection: Raw data was collected from various Indian university websites, educational portals, and visa information sites using the WebScrape Chrome extension.
2. Content Extraction: The pipeline used the trafilatura library to extract the main textual content from raw HTML, discarding irrelevant boilerplate.
3. Semantic Chunking: An LLM was used to perform a "triage" step, segmenting the cleaned text into logical, topic-based chunks (e.g., 'admissions', 'fee_structure').
4. Schema-Guided Structuring: Each chunk was then processed by an LLM with a topic-specific prompt and a corresponding JSON schema, transforming the unstructured text into a highly structured knowledge base.

Part 2: The Q&A Forge (Dataset Generation)

1. Context-Aware Generation: The structured JSON files from Part 1 were used as the full context for an LLM.
2. Instruction Prompting: A master prompt (qa_generation_prompt.md) guided the LLM to generate relevant Q&A pairs from the perspective of a Bangladeshi student, including the answer and all relevant metadata fields.

Part 3: The Quality Assurance Forge

1. Deduplication: A final pipeline step was performed to ensure the quality of the dataset. This stage normalized all questions (by lowercasing and removing punctuation) to identify and remove semantic duplicates.
2. Final Output: The result of this process is the clean, deduplicated dataset.jsonl file.

Intended Use

This dataset is intended for fine-tuning instruction-based LLMs (such as Mistral, Llama, or Gemma) to create a specialized chatbot or AI assistant. The goal is to build a model that can answer questions about Indian universities for Bangladeshi students with a high degree of accuracy and cultural understanding.

Limitations

• The information is based on data scraped from public websites at a specific point in time and may become outdated.
• While the pipeline is designed to be comprehensive, it may not cover every possible topic or university.
• The dataset is highly specialized and may not be suitable for general-purpose chatbot training.

Citation

If you use this dataset, please cite it as follows:

@misc{md_millat_hosen_2025,
  author       = {MD Millat Hosen and Md Moudud Ahmed Misil and Dr. Rohit Kumar Sachan},
  title        = {indian_university_guidance_for_bangladeshi_students (Revision 5c41804)},
  year         = {2025},
  url          = {https://huggingface.co/datasets/millat/indian_university_guidance_for_bangladeshi_students},
  doi          = {10.57967/hf/6295},
  publisher    = {Hugging Face}
}

License

This dataset is released under the MIT License.