Insurance Assist - Inba

Inspiration

We’ve all been there. You pay your premiums for years, but when you’re finally discharged from the hospital, the insurance company hands you a bill full of deductions you don't understand. I’ve lived this. Last year, I was hit with a $300 deduction for something called 'monitoring charges.' The TPA told me they were 'non-medical.' I tried to fight it, but I was buried in jargon. The middlemen weren't there to help me; they were there to protect the bottom line. I knew there was a better way. I used Large Language Models to scan the fine print. I found exactly where the hospital had double-billed me, used IRDAI guidelines to prove it, and won my dispute.

What it does

Built on DigitalOcean’s Gradient AI platform, Inba is your personal insurance advocate. It translates complex policy language into plain English and identifies wrongful deductions in seconds

How we built it

The Brain: We used Mistral Nemo via DigitalOcean’s Serverless Inference for its high reasoning capabilities and efficiency in handling complex legal/medical text.

The Knowledge Base: We implemented Retrieval-Augmented Generation (RAG) by indexing a specialized knowledge base containing the latest IRDAI Master Circulars (2024-2025) and standard insurance policy frameworks. This ensures Inba’s advice is legally grounded, not just "hallucinated."

Challenges we ran into

The biggest hurdle was terminology obfuscation. TPAs often use vague terms like "monitoring charges" or "administration fees" to categorize medical expenses as "non-medical." Mapping these arbitrary hospital billing descriptions to standardized IRDAI non-payable lists required fine-tuning our agent's instructions.

Accomplishments that we're proud of

The Personal Win: The logic behind Inba isn't theoretical—it’s battle-tested. I personally used these LLM-driven insights to dispute a $300 (₹25,000) deduction on my own claim. By identifying where the hospital had double-billed me for services already included in the room rent, I forced a re-evaluation and won.

Jargon Translation: We successfully turned "dense legalese" into "actionable empathy." Inba doesn't just quote rules; it tells the user exactly why their money was taken and how to ask for it back.

Regulatory Accuracy: Our RAG system specifically cites the May 2024 IRDAI Master Circular, giving users the exact authority they need to challenge a TPA's denial.

What we learned

We discovered that the "middleman" system thrives on information asymmetry. Most people give up on disputes because they don't have the time to research 100-page policy documents. We learned that LLMs are the ultimate "equalizers" in this space. They can spot "double billing" (e.g., charging for a pulse oximeter when it's already part of the nursing charges) which is a primary source of revenue leakage for patients. We also realized that the regulator (IRDAI) is increasingly pro-consumer, but consumers simply lack the tools to invoke those protections.

What's next for Insurance Assist - Inba

Scraping from IRDAI Circulars Scraping from IRDAI Circulars to update the knowledge base on regular basis. Multilingual Awareness Inba currently responds only in English but could be extended to other languages in the future. Policy upload Allow users to upload their policy documents to the knowledge base to get customized responses based on their policy. Whatsapp Integration Inba could be integrated with Whatsapp to provide a more seamless experience for users.

Built With

• digitaloceangradientai
• github