Kastle: Scaling AI Agents in Regulated Banking Operations
Y Combinatorgo watch the original →
the gist
Kastle automates high-volume back-office banking tasks like mortgage servicing and collections by deploying autonomous AI agents that integrate with existing legacy systems without requiring a full infrastructure overhaul.
Operational Strategy for Regulated Environments
Kastle addresses the friction banks face when attempting to scale operations while constrained by legacy core systems. Rather than attempting a multi-year rip-and-replace of core infrastructure, the company deploys AI agents that function as digital employees. These agents handle high-volume, labor-intensive tasks such as mortgage servicing, collections, and consumer lending workflows. By focusing on operational automation, the platform allows banks to increase loan and deposit volume without linearly scaling their back-office headcount.
Transitioning from Pivot to Production
After failing to gain traction with two previous ideas—real estate assistance and AI-driven SDR tools—the founders identified a specific, high-pain bottleneck in mortgage servicing: 10 million Americans still make monthly mortgage payments via phone calls handled by human agents.
- Problem Discovery: The founders spent two weeks conducting cold outreach at industry conferences, specifically targeting mortgage lenders to identify operational pain points that were considered existential risks.
- Rapid Prototyping: To secure their first major client, the team moved their entire company to the client's office in Phoenix, Arizona, and remained on-site until the system was fully live and operational.
- Reliability Engineering: Because the agents operate in a highly regulated financial environment, the team built a proprietary layer over LLMs to enforce determinism and compliance, ensuring that non-deterministic model outputs do not violate financial regulations.
Scaling and Future Roadmap
Kastle has processed over $2 billion in transactions to date, growing from $10 million in volume every two months to $10 million daily. The company is currently expanding its agent capabilities beyond mortgage servicing into broader consumer lending, including auto loans, credit cards, and personal lines of credit. The long-term objective is to position their agents as co-workers that handle repetitive tasks, thereby freeing human staff to focus on complex workflows requiring subjective judgment.