Public Sector / Legal AI
Legal Reasoning AI for Tax Adjudication
Client A chartered accountancy and tax practice, India
Legal AIDomain-Adapted ModelAgentic WorkflowGovTech
With a chartered accountancy and tax practice, Jungroo AI built a legal reasoning system for tax adjudication. It takes raw taxpayer data and produces a structured, legally reasoned draft order, linking each finding to the provision behind it. Drafting time fell from 48 to 72 hours per notice to about 30 minutes.
The challenge
Issuing a single adjudication order took 48 to 72 hours of manual work: reading the documents, cross-checking provisions that change often, and studying past case law. Different officers could reach different conclusions on similar facts, which weakens an order when it is challenged.
The solution
Jungroo AI built a legal reasoning system with four parts:
- Domain-Adapted Model: Continued pre-training on the full body of statutory provisions and relevant precedents, so the model reasons in the language of tax law.
- Agentic Drafting Workflow: An agent takes each case from raw data to a structured draft order, step by step.
- Provision Mapping: Every flagged default is linked to the specific statutory provision it relates to.
- Consistent Reasoning: One reasoning template is applied across offices, reducing differences between officers.
The impact
Impact
| Measure | Before | After |
|---|---|---|
| Drafting time per notice | 48 to 72 hours | About 30 minutes |
| Consistency | Depends on the officer | One template, aligned to the statute |
| Where the data lives | Paper files and local systems | Private, inside the client's environment |
| Audit trail | Manual review | Every step of the reasoning traceable |