Retrieval-Augmented Reasoning for Chartered Accountancy

TL;DR

This paper introduces CA-ThinkFlow, a retrieval-augmented reasoning framework for Indian Chartered Accountancy tasks, achieving competitive performance with a parameter-efficient model while highlighting limitations in complex regulatory reasoning.

Contribution

The paper presents CA-ThinkFlow, a novel retrieval-augmented generation system that operates efficiently with a 14B-parameter model for complex CA tasks, matching large proprietary models in performance.

Findings

Achieves 68.75% SRC on CA-Ben benchmark, comparable to GPT-4o and Claude 3.5 Sonnet.

Uses a 14B-parameter, 4-bit-quantized model with retrieval and layout-aware extraction.

Shows high efficiency but struggles with complex regulatory texts.

Abstract

The inception of Large Language Models (LLMs) has catalyzed AI adoption in the finance sector, yet their reliability in complex, jurisdiction-specific tasks like Indian Chartered Accountancy (CA) remains limited. The models display difficulty in executing numerical tasks which require multiple steps while also needing advanced knowledge about legal regulations and the method of scaling their operations is not feasible in settings which have limited access to resources. We present CA-ThinkFlow as a parameter-efficient Retrieval-Augmented Generation (RAG) framework which operates with a 14B, 4-bit-quantized reasoning model, 14B-DeepSeek-R1, and a layout-aware Docling extraction system which maintains document structure during extraction. CA-ThinkFlow uses a basic RAG method which automatically adds retrieved information into the prompt, while it depends on the model's built-in Chain-of-Thought (CoT) functions to create context and produce correct answers. The system we developed system operates at performance levels which match large proprietary models when we tested it on the multi-level CA-Ben benchmark, achieving Scholastic Reliability Coefficient (SRC) results which equal 68.75\% of GPT-4o and Claude 3.5 Sonnet. The framework shows high efficiency and strength in handling parameters, but essential reasoning abilities fail to process complex regulatory texts which exist in fields such as Taxation.