sabin1234/Future_Education_Nepal_MBBS_FAQ_Dataset

Future Education Nepal — MBBS FAQ Dataset (Nepali)

Overview

This dataset (future_education_nepal_mbbs_faq_nepali_sharegpt.jsonl) is a collection of 38 question-answering conversation pairs in Nepali, covering frequently asked questions about studying MBBS (medicine) in Nepal — primarily aimed at Indian students considering Nepal as a study destination. Each record is a single-turn human↔gpt exchange in ShareGPT-style format: a Nepali-language question about MBBS admissions, fees, recognition, or student life, followed by a short, factual Nepali-language answer.

The source is labeled "Future Education - MBBS in Nepal", and unlike the two other datasets in this series, this one is marked generation_type: human-curated and condition: real-world (rather than real) — suggesting the Q&A content was curated by humans from real informational material about MBBS study in Nepal, then formatted into an instruction-tuning conversation structure.

• File format: JSON Lines (.jsonl), one JSON object per line
• Total records: 38
• Source: Future Education - MBBS in Nepal (repo: Future Education)
• License: Apache-2.0 (permissive tier)
• Language: Nepali (ne / ISO 639-3 npi), Devanagari script (Deva)
• Task type: question-answering
• Generation type: human-curated
• Condition: real-world
• Source revision: v1

File / Record Structure

Every line is a JSON object with the following top-level fields:

Field	Description	Observed value(s)
id	Unique record identifier	38 distinct sg_<32-char hex hash> strings, e.g. sg_09138666ec5037d5cc53e4dcff3baea7
conversations	List of 2 turns: human (question) then gpt (answer)	always exactly 2 turns
source	Dataset source label	Future Education - MBBS in Nepal
source_name	Internal dataset name	Future Education - MBBS in Nepal
source_repo	Source repository name	Future Education
source_config	Config/subset name	default
source_split	Data split	train
source_revision	Source version tag	v1
source_row_id	Row ID in the original source	same hash as id, suffixed with :1 (e.g. sg_09138666...baea7:1)
language	Language code	ne
language_code	ISO 639-3 code	npi
script	Writing system	Deva (Devanagari)
license	License	Apache-2.0
license_tier	License permissiveness tier	permissive
task_type	Task category	question-answering
generation_type	How the data was produced	human-curated
condition	Data condition/type flag	real-world
url	Source URL	empty string for all records
metadata_json	Stringified JSON with generation-level metadata (see below)	—

metadata_json sub-fields (per record)

Field	Description	Observed value
generation_domain	Broad subject domain	शिक्षा ("Education")
generation_category	Category	नेपालमा एमबीबीएस ("MBBS in Nepal")
generation_sub_domain	Sub-topic	एमबीबीएस भर्ना तथा अध्ययन ("MBBS admission and study")
behavior	Expected model behavior	छोटो र जानकारीमूलक उत्तर दिने ("Give a short and informative answer")
content_language	Content language (Nepali label)	नेपाली
content_script	Content script (Nepali label)	देवनागरी
english_content_allowed	Whether English content is permitted	false

This dataset's metadata schema is leaner than the HIB health insurance FAQ dataset — it has no question_type, question_length, response_length, maximum_response_sentences, minimum_reasoning_dimensions, or minimum_complexity_score fields, and no behavior_definition field either (only the short behavior label).

Distribution Analysis

Behavior Distribution

Only one behavior type appears across the entire dataset:

Behavior	Count	%
छोटो र जानकारीमूलक उत्तर दिने ("Give a short and informative answer")	38	100%

Domain / Category / Sub-domain Distribution

Field	Value	Count
Domain	शिक्षा ("Education")	38/38
Category	नेपालमा एमबीबीएस ("MBBS in Nepal")	38/38
Sub-domain	एमबीबीएस भर्ना तथा अध्ययन ("MBBS admission and study")	38/38

This is a single-domain, single-category, single-sub-domain dataset — every field-level metadata value is constant. All topical variety lives in the question text itself. Reading through the 38 questions, they cluster into these practical themes (a single question sometimes touches more than one):

1. Recognition & degree validity (~7 questions): whether an MBBS from Nepal is valid in India, USA/UK/Australia, WHO recognition of Nepal, comparing Nepal's MBBS to India's private MBBS, comparing Nepal vs. Russia for MBBS
2. Admissions & eligibility (~9 questions): NEET requirement and minimum score, entrance exam process, required documents, minimum age, admission process for 2026, what happens if a student fails an exam, whether local-language learning is required
3. Fees & duration (~2 questions): total MBBS fee in Nepal, program duration
4. Universities & institutions (~3 questions): number of NMC-approved universities, list of major NMC-approved universities, which city is best for MBBS study
5. Student life & logistics for Indian students (~9 questions): cost of living, student life experience, safety, food situation, hostel facilities, weather/climate adaptation, whether parents can visit, visa-free entry for Indian citizens
6. Post-graduation & career (~6 questions): FMGE pass percentage for Nepal graduates, best FMGE prep strategy, pursuing MD/MS in India afterward, career options after MBBS, internship opportunities
7. Key terms/exams explained (~4 questions): what is NMC, what is FMGE and its difficulty, what is NEXT and whether it replaces FMGE, medium of instruction (English)
8. Scale/statistics (~1 question): current number of Indian MBBS students in Nepal

Content Language / Script Distribution

Field	Value	Count	%
content_language	नेपाली (Nepali)	38	100%
content_script	देवनागरी (Devanagari)	38	100%
english_content_allowed	false	38	100%

Purely monolingual Nepali content in Devanagari script — no English or code-switching permitted, even though the subject matter (Indian students studying MBBS abroad) is inherently cross-border.

Task Type / Generation Type / Condition Distribution

Field	Value	Count	%
task_type	question-answering	38	100%
generation_type	human-curated	38	100%
condition	real-world	38	100%

Notably, this differs from the other two datasets in the series (SAC Nepal health/fitness FAQ and HIB health insurance FAQ), which both use task_type: instruction-following and generation_type/condition: real. This dataset instead uses task_type: question-answering and generation_type: human-curated / condition: real-world — indicating a slightly different data-generation pipeline or labeling convention was used for this source.

Other Constant Fields

The following fields are identical across all 38 records (no variation to report): source, source_name, source_repo, source_config, source_split, source_revision, language, language_code, script, license, license_tier, and url (always empty).

Conversation Structure

Metric	Value
Turns per conversation	Always 2 (1 human + 1 gpt)
Roles used	human, gpt
Human turns	38
GPT turns	38

Single-turn instruction/response format — no multi-turn dialogue.

ID Structure

• id: a sg_ prefix followed by a 32-character hex hash (likely an MD5-style hash), all 38 unique
• source_row_id: identical hash to id, with a :1 suffix appended (e.g. sg_09138666ec5037d5cc53e4dcff3baea7:1) — the :1 likely denotes "turn/segment 1" from the original ShareGPT-format source record

Text Length Statistics

Metric	Min	Max	Average
Question length (characters)	30	88	~54
Answer length (characters)	123	441	~261

Questions are short, single-sentence prompts. Answers are moderately detailed — averaging 261 characters, roughly in between the SAC Nepal fitness dataset (121 chars) and the HIB health insurance dataset (~327 chars) — consistent with the behavior label calling for "short and informative" answers.

Summary

This dataset is a narrow, single-purpose Q&A corpus: 38 short Nepali-language question/answer pairs, all about MBBS study in Nepal from the perspective of prospective (largely Indian) medical students. Every record shares the same domain (Education → MBBS in Nepal), the same sub-domain (MBBS admission and study), and the same behavior instruction ("short, informative answer"). There is no variation across domain, category, license, task type, or content-language settings — the only real variety is in the specific sub-topic of each question, spanning degree recognition, admissions/eligibility, fees, universities, student life/logistics, post-graduation career paths, key exam/regulatory terms (NMC, FMGE, NEXT), and enrollment statistics.

Compared to its sibling datasets:

• SAC Nepal (health/fitness): broader health domain, instruction-following/real labeling, shortest answers (~121 chars avg)
• HIB (health insurance): narrower "Health Insurance" domain, richest metadata (question/response length + complexity constraints), longest answers (~327 chars avg)
• Future Education (MBBS): Education domain, question-answering/human-curated/real-world labeling, leanest metadata schema, mid-length answers (~261 chars avg)

This kind of dataset is well suited for fine-tuning or evaluating a language model's ability to answer study-abroad/medical-education FAQs accurately and concisely in Nepali, or as a domain-specific supplement (education vertical) alongside the health-focused datasets already reviewed in this series.