Where We’re Going, Where We’ve Been
2024 was a banner year for OpsLab.
Coming off the heels of our seed round investment in late 2023, we made the strategic decision to acquire a leading provider of operational squadron planning software to the U.S. Air Force. We expanded our capabilities beyond just flight scheduling for fixed-wing assets, and began servicing unmanned aircraft and managing air ranges. We refreshed our brand, announced our strategic advisors, and broke into the international market. We’re now operational in 60 squadrons across the U.S. Air Force with more than 2,500 end users.
As we think about how we take 2024’s momentum into 2025, it’s those thousands of end users that we are acutely focused on. Maintaining such a strong userbase means the foundations of our product are strong, and now, as we turn to 2025, we’re aiming to make our offering even better by adding new AI capabilities in very short order. Here are a few of the core AI capabilities we're working on in 2025:
Predictive Readiness. Nearly every product conversation we have with our U.S. military customers comes back to readiness. It is critical that we not only maintain, but increase readiness to best prepare for future conflict. Though the OpsLab product suite has already proven to increase readiness, we want to go further by enabling commanders to actually predict readiness levels. To do this, we’re using an AI technique called supervised machine learning to forecast critical customer metrics, particularly focusing on student throughput. This system not only provides predictions of expected pilot throughput for the year but also offers interactive features allowing customers to explore different scenarios. Users can adjust variables such as the number of available aircraft or instructors and immediately see how these changes might affect overall operational readiness. The system has potential for further development into a comprehensive communication engine that could provide specific recommended actions and conduct what-if analyses for customers, enabling more informed decision-making and resource allocation.
Self Calibrating Automation. Next, we’re using reinforcement learning (RL) to make better automated decisions. With this approach, we set up the squadron scheduling problem as a Markov Decision Process (MDP). In the MDP, the environment is a squadron’s operational status including key metrics, the state space is the current schedule including historical schedule data, and the action is the assignment of a pilot to a mission. After training an RL agent using the wealth of historical squadron operations data uniquely available to OpsLab, we create a stochastic policy that provides a probability distribution over the selection of pilots for specific missions on specific dates and times. This policy can be used to rapidly generate multiple high-quality schedules, each of which can then be cleaned up by our baseline optimization model and presented to the user for review. The main benefit of RL is the incorporation of user feedback into the training loop: feedback given about the schedule by users is recorded and added to offline data used in retraining the RL policy. This enables our scheduling system to automatically calibrate to evolving user requirements, and allows for rapid decision making while maintaining flexibility in schedule management.
Natural Language Interaction. Finally, we’re leveraging large language models (LLM) to create an intuitive interface for squadron performance assessment. This development was made possible by the extensive foundational work in automation and data architecture. The system allows users to ask questions in natural language about their squadron's performance and receive clear, conversational responses. Behind the scenes, it translates these natural language queries into database commands, retrieves the relevant information, and converts it back into easily understood responses. While implementing a chatbot interface might seem straightforward using open-source models with fine-tuning, the real innovation lies in the underlying infrastructure. Without the carefully designed database architecture, high-quality data collection systems, and overall system integration, such a capability would be impossible to implement effectively.
These new AI capabilities represent an exciting evolution in our approach, but it's important to recognize that our foundation in Operations Research is what made this possible. The rigor and precision of OR gave us the deep understanding of squadron operations and the robust data architecture needed to even consider these AI implementations.
But as we truly think about maximizing our user experience, we see tremendous potential in combining the mathematical precision of OR with the adaptability and intuitive interfaces of AI. This isn't about replacing one approach with another – it's about creating a powerful synergy where each technology enhances the other.
We're particularly excited about weaving these features into our existing tools, enhancing what already works while adding new dimensions of functionality. At the end of the day, our focus remains unchanged: we want to reduce the time and cost burdens of squadron scheduling and planning to improve operational readiness. By bringing together the best of OR and AI, we're better positioned than ever to deliver on that mission.
