README.md

---
license: cc-by-nc-4.0
task_categories:
  - question-answering
language:
  - en
tags:
  - finance
  - indian-finance
  - benchmark
  - financial-qa
  - llm-evaluation
pretty_name: IndFin-Bench
size_categories:
  - n<1K
---

# IndFin-Bench: A Benchmark Grounded in Indian Financial Filings

**IndFin-Bench** is a benchmark of 100 hand-curated questions sourced from the corporate filings of Indian listed companies. It is designed to evaluate how accurately LLMs can retrieve and reason over India-specific financial data.

Existing financial benchmarks like [FinBen](https://github.com/The-FinAI/FinBen) and [FinQA](https://github.com/czyssrs/FinQA) are built on US market data — SEC filings, 10-K reports, and earnings calls from American corporations. **IndFin-Bench fills this gap for the Indian market**, covering the Ind AS accounting landscape, SEBI disclosure formats, and cross-company reasoning over BSE/NSE-listed equities.

---

## Dataset Summary

| | |
|---|---|
| **Questions** | 100 |
| **Companies** | 100+ Indian listed firms |
| **Coverage** | FY23 - FY26 |
| **Source** | Annual reports, quarterly filings, investor presentations, BSE/NSE disclosures |
| **License** | CC-BY-NC 4.0 (with additional restrictions; see below) |

---

## Schema

| Field | Description |
|---|---|
| `Question_Num` | Unique identifier for the question |
| `Question` | Natural-language financial query grounded in official filings |
| `Complexity_Category` | 4-tier taxonomy from single-company single-fact to multi-company multi-fact |
| `Atomic_Fact` | Verified ground truth answer, sourced directly from official filings |

---

## Complexity Levels

| Category | Questions | What it tests |
|---|---|---|
| Single-Company, Single Fact | 38 | Retrieve one specific figure from one company's filing |
| Single-Company, Multiple Facts | 33 | Retrieve and synthesise multiple data points from a single company |
| Multi-Company, Multiple Facts | 20 | Compare or derive metrics across multiple companies |
| Multi-Company, Single Fact | 9 | Retrieve comparable figures across companies and perform comparison |

---

## Usage

```python
from datasets import load_dataset

ds = load_dataset("CompoundingAI/IndFin-Bench")
```
---

## Ground Truth

Each question has a corresponding **Atomic Fact** — a precise, minimal statement of the correct answer derived directly from official company filings. Every atomic fact has been manually verified against primary filings.

---

## Intended Use & Restrictions

This dataset is released under the **Creative Commons Attribution-NonCommercial 4.0 (CC BY-NC 4.0)** license.

**IndFin-Bench is strictly intended for evaluation and benchmarking of language models on Indian financial data.**

Typical allowed use cases include:
- Evaluating LLM performance on financial question answering
- Benchmarking retrieval-augmented generation (RAG) systems
- Comparing model accuracy across different complexity tiers

**Use of this dataset implies agreement with the following restrictions:**

- The dataset **must not be used for training, fine-tuning, or improving** any machine learning or AI models.
- This includes (but is not limited to):
  - Supervised training  
  - Fine-tuning or instruction tuning  
  - Reinforcement learning (RL / RLHF)  
  - Retrieval-augmented training pipelines  
  - Synthetic data generation derived from this dataset  

- The dataset must be used **only for evaluation purposes and not as a training corpus**.

---
## Citation

If you use IndFin-Bench in your work, please cite:

```
@dataset{compoundingai2026indfinbench,
  title={IndFin-Bench: A Pioneering Benchmark Grounded in Indian Financial Filings},
  author={CompoundingAI},
  year={2026},
  url={https://huggingface.co/datasets/CompoundingAI/IndFin-Bench},
  license={CC-BY-NC-4.0}
}
```

## Links

- [Blog Post](https://compoundingai.in/benchmark)
- [CompoundingAI](https://compoundingai.in)