Terminal X's Eval Agent, running on OpenAI's GPT-6.1 Sol, matches senior investor judgment 88.6% of the time, according to the company. The system catches 25% more deviations from a fund's mandate compared to other models.
TL;DR
- Terminal X's Eval Agent, powered by GPT-6.1 Sol, achieves 88.6% alignment with senior investor judgment.
- The system identifies 25% more mandate deviations and reduces cost per query by 30%.
- Terminal X aims to encode individual investment professionals' judgment, not just firm-wide standards.
What happened
Terminal X, a fintech startup focused on preserving investment firms' decision-making processes, has developed an Eval Agent that evaluates agents' outputs against a 54-criterion rubric. Powered by OpenAI's GPT-6.1 Sol, the Eval Agent matches senior investor judgment 88.6% of the time, outperforming other models by 220 basis points.
The Eval Agent catches about 25% more recommendations that stray from a fund's mandate and runs workflows at approximately 30% lower cost per query than the next-best model. It checks each output for numerical accuracy, narrative consistency, and alignment with the firm's house view.
Terminal X has built 1,700 firm Memories, run over 1 million custom workflows, and reached a weekly active user count equal to 81% of its monthly active users. The company plans to extend firm Memories from desks to individual investment professionals.
Why it matters
For AI/ML developers, this demonstrates the potential of fine-tuning language models for specialized tasks in finance. The Eval Agent's performance highlights the importance of model selection and evaluation in real-world applications.
Startup founders in the fintech and AI spaces can learn from Terminal X's approach to preserving institutional knowledge and improving decision-making processes. The company's focus on encoding individual judgment could inspire similar efforts in other industries.
Tech investors should take note of the significant cost savings and efficiency gains reported by Terminal X's customers. The company's traction, with a high weekly active user count, suggests a strong product-market fit and potential for growth.
Key facts
- Terminal X's Eval Agent matches senior investor judgment 88.6% of the time using GPT-6.1 Sol (Terminal X).
- The system catches 25% more mandate deviations and reduces cost per query by 30% (Terminal X).
- Eval Agent checks outputs against a 54-criterion rubric, with each fail including a one-line reason and correction (Terminal X).
- Terminal X has built 1,700 firm Memories and run over 1 million custom workflows (Terminal X).
- Weekly active users equal 81% of monthly active users, indicating strong engagement (Terminal X).
- One private equity firm reported $2.4 million in annualized savings from automated due diligence and investment memo workflows (Terminal X).
- Terminal X plans to extend firm Memories from desks to individual investment professionals (Terminal X).
- The company's forward-deployed engineers include former finance and investment professionals from JPMorgan and Blackstone (Terminal X).
Context
Terminal X's Capital Memory operating system is designed to address 'memory debt'—the loss of reasoning behind investment decisions when senior investors leave. By encoding each firm's best practices and judgment into agents, the company aims to make investment processes more measurable, auditable, and reusable.
The use of OpenAI's GPT-6.1 Sol and GPT-6 Luna models highlights the growing trend of AI adoption in the finance industry. These models handle reasoning tasks and high-volume steps, respectively, demonstrating the versatility of large language models in specialized applications.
Terminal X's approach to preserving institutional knowledge and improving decision-making processes could have implications beyond the finance industry. As AI continues to advance, more companies may seek to encode and reuse the expertise of their most valuable employees.