A developer has introduced itr-wala — an open-source solution that allows filing Income Tax Returns (ITR) in India directly via the terminal, leveraging the capabilities of AI agents such as Claude Code, Codex, and Gemini.
What Happened
The itr-wala project uses a hybrid approach to automate tax processes: AI agents are responsible for extracting data from unstructured documents (such as Form 16 and AIS), while all mathematical calculations are performed by a deterministic Python engine. The tool supports calculations for both the old and new Indian tax regimes for the 2026-27 tax year and includes a testing system to verify accuracy.
Context
Traditional LLMs often make errors in arithmetic calculations ("hallucinations"), making them unreliable for critical financial tasks. The project solves this problem by separating responsibility between the neural network (acting as an interface for text processing) and programmatic code (acting as the computational core).
Why It Matters for the Industry
The project demonstrates a mature architectural pattern for creating production-ready AI agents in high-risk and regulated industries. The use of the "LLM for unstructured data + deterministic code for logic" model could become an industry standard when developing agents for finance, compliance, and auditing.
Why It Matters for Users
For users, this is an example of how complex and routine financial procedures can be automated in a terminal-first manner, minimizing the risk of human error and model errors through integration with a reliable computational engine.
Sources
Author
Look at AI, Editorial Team