Sample Dataset Generator

[ms-shashank]

Current file: app/src/lib/tools/sample-dataset-generator.ts
Current model: llama-3.3-70b
Current approach: Single prompt asking for sample data in CSV/JSON/SQL format. No schema enforcement, no data validation, no consistency checks.

Problems with current approach:

• Generated data often has inconsistent column counts or malformed rows.
• Edge cases (nulls, boundary values) are unreliable.
• Output format may not parse correctly (e.g., invalid JSON, broken CSV quoting).
• No referential integrity between related entities.

Upgrade plan:

Step	Agent	Action
1	Schema Designer	Analyze the user's description and generate a formal schema: column names, data types, constraints, relationships between tables (if any).
2	Data Generator	Generate realistic records following the schema. Include edge cases per the plan.
3	Format Compiler	Programmatic: Convert the generated data into the requested format (CSV, JSON, SQL, Markdown table) using Python serializers to guarantee valid output.
4	Validation	Programmatic: Validate output format (parse JSON, validate CSV row counts, check SQL syntax). If invalid, feed errors back to Step 2 for regeneration. Max 2 retries.

• You are free to enhance the agents stacks in the above plan layout, the above one is just for reference. You can enhance more if needed.

Model suggestions to start with:

• Step 1: Try deepseek-v3.2 or llama-4-maverick-17b for schema design (structured but not complex).
• Step 2: Try llama-3.3-70b or qwen-3-32b for data generation. Also try minimax-m2.5 which handles structured outputs well.

Model Selection Guidance

• You are free to pick any model from the Oxlo catalog based on your own testing and evaluation.
• The Models suggestions above, not mandates. Try them first, and if they do not meet the accuracy target, experiment with alternatives.

Compare against: GPT 5.3 Thinking & Claude Sonnet 4.6 Thinking.

Acceptance criteria:

• Output format must be parseable in 100% of cases (valid JSON, valid CSV, valid SQL).
• Generated data includes at least 2 edge cases per dataset (null values, boundary values, empty strings).
• Row count matches the user's requested count.
• Overall accuracy at 80%+.