When you ask TaxMarg a tax question, the answer does not come from a language model's memory. It comes from a search across our indexed database of Indian tax legislation, rules, circulars, and notifications, with every citation verified against the source text.
Here is how it works at a high level, without revealing the proprietary details.
Step 1: Understanding Your Question
When you type a question like "What is the TDS rate on payment to a contractor under the new IT Act?", TaxMarg first understands what you are actually asking:
- Subject: TDS on contractor payments
- Applicable law: Income-tax Act, 2025 (since you said "new IT Act")
- What you need: The rate, the threshold, and the relevant section number
This is not keyword matching. The system understands natural language queries, including colloquial terms CAs commonly use. "What's the 194C rate?" and "TDS on contracts under new Act" are understood as the same question.
Step 2: Searching Across Databases
TaxMarg does not search one document. It searches across multiple databases simultaneously:
- The full text of the Income-tax Act, 2025
- The Income-tax Act, 1961 (for cross-referencing)
- Income-tax Rules, 2026
- CBDT Circulars and Notifications
- GST legislation (CGST Act, IGST Act, GST Rules)
- GST Circulars and Notifications
- Double Taxation Avoidance Agreements
- Finance Acts and amendments
- Allied tax acts
The search is not a simple keyword search. It uses a combination of semantic understanding (finding provisions that are about the same concept, even if they use different words) and precise text matching (finding exact section references).
Step 3: Verifying Citations
This is where TaxMarg is fundamentally different from ChatGPT or any general-purpose AI.
Before including any citation in the response, TaxMarg verifies:
- Does the section exist? Section 393 of the IT Act 2025 is a real section. "Section 394A(2)(b)(iii)" might not be.
- Is the section relevant? The section must actually deal with the topic being discussed, not just contain a related keyword.
- Is it current? If a circular has been superseded by a later circular, the older one should not be cited as current authority.
- Is the text accurate? The quoted text from the section must match the actual legislative text.
Only verified citations make it into the response.
Step 4: Generating the Answer
With the relevant, verified provisions in hand, TaxMarg generates a structured answer:
- Direct answer to the question (e.g., "The TDS rate on payment to a contractor is 1% for individuals/HUFs and 2% for others, under Section 393 of the IT Act 2025")
- Cited provisions with the exact section text
- Cross-references to the old Act (e.g., "This corresponds to Section 194C of the IT Act 1961")
- Related provisions that might be relevant (e.g., threshold limits, exemptions)
- Practical notes (e.g., "Note: No TDS is required if the single payment does not exceed Rs. 30,000 and total payments during the year do not exceed Rs. 1,00,000")
What We Do Not Do
We are transparent about the boundaries:
- We do not provide legal advice. TaxMarg is a research tool. The answer is a summary of the law, not a recommendation for a specific client situation.
- We do not guarantee completeness. Tax law is vast, and there may be relevant provisions we have not yet indexed (particularly older tribunal orders).
- We do not replace a CA's judgement. The CA must evaluate whether the retrieved provisions apply to their client's specific facts.
Why This Approach Works
The key insight is simple: search and verify, do not generate and hope.
General-purpose AI models generate answers from patterns learned during training. This works well for general knowledge but fails for precise legal citations where accuracy is non-negotiable.
TaxMarg's approach ensures that every section number, every rate, every threshold in the response can be traced back to the actual legislation. That is the standard CAs need, and that is what we deliver.