Overview

Assort Health is on a mission to make exceptional healthcare accessible anytime, anywhere. The company’s platform leverages generative AI to automate call center operations in medical and dental practices, dramatically simplifying patient scheduling and triage.  Patient registration, scheduling and other tasks are streamlined with zero human intervention, enabled by seamless, deep integrations with EHR (Electronic Health Record) and PM (Practice Management) systems.

"We're the ultimate call center operator. Our AI agent never calls in sick, never forgets a rule, and is always online." — Jeffery Liu, Founder and co-CEO, Assort Health

However, delivering this experience in real-time comes with formidable challenges. Each healthcare practice has unique, highly complex scheduling rules—sometimes documented in binders full of custom logic. At the same time, call center staffing shortages and turnover mean that reliability and consistency are essential. Assort’s voice-based AI agent must handle thousands of patients speaking in countless different ways, all while maintaining low latency, high accuracy, and strict compliance.

Assort Health’s generative AI agents understand caller needs, resolving patient inquiries without human intervention.

Business Challenge: Ensuring High-Quality, Low-Latency AI Interactions

While Assort’s generative AI agent promised a transformative patient experience, developing and maintaining it proved difficult. The main challenges included:

1. Complex, Custom Scheduling Logic:
   Healthcare appointments aren’t as simple as booking a table at a restaurant. Patients must provide demographic details, insurance information, chief complaints, and more. Practices have intricate logic—often tens of pages of rules—determining which physician a patient should see and when. Ensuring the AI agent learned and applied these rules accurately, every time, was critical.

2. Latency Sensitivity and Real-Time Performance:
   In a live voice interaction, even small delays can frustrate users. A synchronous agent must respond quickly, leaving no room for slow or manual debugging processes. Without proper observability, identifying and fixing lag-inducing logic or delays in Large Language Model (LLM) calls could be slow and error-prone.

3. Troubleshooting and Debugging Complexity:
   When an issue arose—such as the agent overwhelming patients by listing too many time slots (“9 AM, 10 AM, 11 AM…3 PM”) or misinterpreting a subscriber ID—engineers needed to reconstruct entire conversations using python scripts to understand what happened. Then, debugging a single issue could take an additional 15–20 minutes of manual effort. In a system serving over 160,000 patients, those delays added up fast, slowing the pace of improvements and risking decreased patient satisfaction.

4. Continuous Fine-Tuning for Diverse User Interactions:
   Patients use a wide range of communication styles, from phonetic spellings to colloquialisms. Ensuring the agent understood these variations required repeated testing and prompt refinements. Without a rapid, low-friction way to replay scenarios and make incremental changes, it was difficult to quickly improve the agent’s responses.

Solution: Implementing Log10 for Observability and Rapid Iteration

Assort turned to Log10 to address these core challenges. Log10’s robust logging, tagging, and replay capabilities allowed the team to precisely observe and refine every element of the patient-agent interaction:

1. Immediate Insight into Conversations:
   Log10 tags each conversation with a unique ID, enabling engineers to locate and review the exact patient-agent exchange at any time. Instead of manually reconstructing conversations, engineers could open the Log10 dashboard and instantly see a detailed, chronological record of every LLM call and conversation.

2. Rapid Debugging and Reduced Engineer Overhead:
   Previously, resolving a single issue required up to 20 minutes of manual effort. With Log10, the same troubleshooting process became a matter of seconds or minutes. Engineers could replay conversations, identify the root cause of an error, adjust prompts, and retest—without writing custom scripts or guesswork.

3. Iterative Prompt Refinement and Accuracy Improvements:
   By enabling quick playback of interactions, Log10 helped Assort fine-tune the AI agent’s behavior. For example, if the agent overwhelmed patients with too many appointment times, engineers could adjust the prompt, replay the scenario, and confirm the improved experience immediately. Similarly, for challenges like interpreting subscriber IDs or handling unexpected phrasing, Log10’s replay feature made it simple to test multiple strategies and converge on the best solution.

4. Cost Tracking and Metrics Visibility:
   Log10’s dashboard provided valuable metrics—like how long each LLM call took—to help the team understand the system’s performance profile. While Log10 itself did not reduce latency, its metrics allowed Assort to identify performance bottlenecks and confirm that changes to prompts and logic didn’t introduce new delays.

Log10’s robust logging, tagging, and replay capabilities allowed the team to precisely observe every element of the patient-agent interaction and rapidly improve prompts.

“With Log10, everything is tagged with a conversation ID, allowing us to quickly see exactly what happened during any interaction.” – Jeffery Liu, Founder and co-CEO, Assort Health

Results and Impact

• 🚀 5x Faster Debugging: With Log10, the time spent debugging issues shrank dramatically, improving the speed of troubleshooting and prompt engineering by up to five times. Faster iteration meant Assort could deliver improvements to patients and practices more quickly, maintaining a seamless, helpful experience.

• 😊 Better Patient Experience: By rapidly identifying pain points—such as overwhelming the patient with too many options or confusion over subscriber ID entry—Assort could iterate its prompts and logic to create a more natural, intuitive conversation. The result was a smoother patient journey and more accurate provider matching.

• 📈 Improved Observability for Growth: As Assort prepares to serve millions of patients, having granular insight into every interaction is critical. Log10 provides the necessary observability to maintain and enhance quality at scale. The team can easily ensure that prompt adjustments and logic changes do not introduce regressions, protecting the user experience as the platform grows.

• 🔮 Future Development and Reliability
  Assort plans to use Log10’s real-time latency metrics and proactive tagging to scale efficiently while maintaining reliability. By monitoring response times and automatically switching to faster LLM models during slowdowns, they will ensure seamless interactions. Proactive tagging will identify patterns, such as recurring issues in insurance collection or provider triage, triggering alerts for rapid intervention. This approach allows Assort to quickly address issues and uphold excellence in patient care.

“The observability Log10 offers is crucial. We can review every single call, understand the context, and iterate faster—this level of insight is a game-changer.”  – Jeffery Liu, Founder and co-CEO, Assort Health

Conclusion

For Assort Health, Log10 wasn’t just a convenience—it was a transformative tool that brought clarity, speed, and confidence to the highly complex task of automating healthcare call centers with generative AI. By providing deep observability, rapid debugging capabilities, and a straightforward interface for scenario replay, Log10 enabled Assort to focus on what matters most: delivering exceptional, accessible healthcare experiences to every patient who calls.

“Log10 enables us to quickly identify areas for improvement, fine-tune our AI agent’s behavior, and ensure patients have a smoother, more intuitive experience.” – Jeffery Liu, Founder and co-CEO, Assort Health