Our work on Finance AI has been accepted in FinNLP@EMNLP 2025, we present "Enhancing Financial RAG with Agentic AI and Multi-HyDE"

💡 In finance, a single AI hallucination can cost millions.

So how do we make retrieval-augmented generation (RAG) more trustworthy?

In our recently accepted work at FinNLP, EMNLP 2025, we present “Enhancing Financial RAG with Agentic AI and Multi-HyDE.”

🔎 What we did:

• Used Multi-HyDE to generate diverse queries for stronger retrieval coverage.
• Designed Agentic AI workflows for efficient, multi-step reasoning.
• Added domain-specific tools (keyword + table retrieval) to handle structured financial data.

📊 What we found:

• +11% accuracy improvement on financial QA benchmarks.
• ~15% reduction in hallucinations, a big step for reliability.

Paper link: https://arxiv.org/abs/2509.16369

Would love to hear from others working on RAG, finance, and trustworthy AI — where do you see the next bottleneck?

Huge kudos to the team Ryan George, Jayden Koshy Joe, Harshith M R, Rahul Vimalkanth, Hrushikesh Kant, Sudharshan Suresh, Sachin Sundar and J VIJAYAVALLABH

Looking forward to collaborations and feedback. Hope to see you in Suzhou, China in November!