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.