Biz-on GPT: High-Precision Tax Consulting
The Challenge
UAE corporate tax laws are complex and multi-layered. Businesses need immediate answers that cite official federal laws, not just AI-generated guesses.
The Result
Built a trusted advisory tool that cross-references user queries against a massive database of Federal Laws and Cabinet Decisions.
The Challenge: When Hallucinations Are Not an Option
In tax consulting, being “mostly right” is the same as being wrong. A founder navigating UAE Corporate Tax doesn’t just need an answer—they need the specific article of the Federal Law that backs it up. Generic AI models often hallucinate legal details, which can lead to expensive compliance errors.
The goal was to build a system where every word of the AI’s response is grounded in official government documentation.
The Hybrid RAG Solution
We developed a specialized Hybrid Retrieval-Augmented Generation (RAG) engine designed specifically for legal text.
1. Vector + Keyword + Graph Search
Instead of relying on a single search method, Biz-on GPT uses three layers of retrieval:
- Vector Search: To understand the intent behind a founder’s question.
- Keyword Search: To ensure specific legal terms and article numbers are never missed.
- Knowledge Graph: To understand the relationships between different laws and amendments.
2. Multi-Step Verification
Before the user sees an answer, the system performs a multi-step check:
- It retrieves relevant document chunks.
- It “fuses” these results using Reciprocal Rank Fusion (RRF) for maximum accuracy.
- It provides direct links to the source PDF, highlighting the exact paragraph used for the answer.
3. Multilingual Native Support
The system handles queries in English, Russian, Polish, and Hindi, allowing international founders to navigate local regulations in their native language while still receiving answers grounded in the original Arabic and English legal texts.
The Impact for the Business
- Instant Compliance: Users get answers in seconds that would normally take hours of manual research or expensive legal consultation.
- High Trust: By citing every source, the platform builds the level of credibility required for a legal-tech product.
- Operational Scalability: The system handles thousands of queries daily, maintaining the same level of precision across every interaction.