Project Story

Inspiration

We were inspired by a gap between the growing use of AI and the reality of high-stakes financial work. IPO prospectus drafting is still highly manual, repetitive, and time-sensitive, yet errors can lead to serious legal and financial consequences. This led us to ask:

Can AI improve efficiency in regulated drafting without compromising reliability and ethics?


What We Built

We designed a retrieval-augmented multi-agent platform that turns prospectus drafting into a structured workflow:

  • Retrieve relevant precedents
  • Guide structured issuer inputs
  • Generate grounded section-level drafts
  • Verify completeness and consistency

Instead of free-form generation, outputs are constrained by both inputs and retrieved documents: $$ y = f(x_{\text{issuer}}, x_{\text{retrieval}}) $$


Challenges

The main challenge was balancing efficiency and trust.
Unconstrained AI is fast but unreliable, while strict controls limit flexibility.

We addressed this by:

  • grounding outputs in real precedents
  • breaking tasks into sections
  • keeping humans in the loop

Another challenge was translating AI ethics into system design, such as traceability, privacy, and accountability.


Reflection

This project taught us that in high-risk domains, AI should not replace professionals, but support them.

The future of AI is not full automation, but responsible collaboration with humans.

Built With

Share this project:

Updates