We were so fortunate to have had a great group of engineering and product interns this summer, who took part in a number of influential projects at Infinitus, including multilingual capabilities for patient-facing calls, a privacy-focused and scalable call analysis solution, and MCP support for Infinitus APIs (which we even put out a press release about!).

But that’s not all. Read on to learn more about the impressive, influential work that our 2025 class of interns contributed to.

Amanda Griffin: Diving deep on AI operator efficiency

Amanda’s 12-week product design internship resulted in a significant and immediate impact on our AI operators’ efficiency. She worked with Myra Orgain to own a major redesign of the operator portal call page, where her deep user research and cross-functional collaboration helped solve key pain points like digital source data correction accuracy and the need to save crucial information during critical moments during calls, such as payor spiels. Amanda delivered a compelling solution that moved directly into the development pipeline, and her designs – including a live transcript message pinning feature – are already providing value to our users.

Vishal Sankar Ram: Saving time for the Engineering team

Intern Vishal teamed up with Youngseo Son and Myra Orgain to work on two projects this summer:

1. Automated IVR navigation LLM improvement pipeline, which focused on automating the creation and refinement of IVR navigation prompts to save significant engineering time. Vishal built a pipeline that automatically generates prompts for new IVR tasks by applying transfer learning from an existing source domain (Claim IVR). This system drastically reduces the manual design time for new prompts. Infinitus is now gradually rolling out LLMIVR navigation using this optimized prompt for benefit verification (major medical) tasks.

2. Cluster-based + RAG IVR navigation model, which involved developing a more advanced model architecture to boost accuracy and simplify maintenance. Vishal designed a model that clusters similar IVR utterances and applies a unique, optimized prompt for each cluster. This was enhanced with Retrieval-Augmented Generation (RAG) to dynamically provide the model with relevant examples and hints during a call. The RAG integration improves the model’s ability to generalize, and allows for rapid performance gains simply by adding new data, removing the need for complex prompt re-engineering.

Arjun Imandar: Bringing interoperability to healthcare with MCP

This summer, Arjun partnered with Shwetha Shinju to work on creating an MCP server that transforms Infinitus’s APIs into a toolkit for AI agents. He also built out our AI agents on Salesforce. These two projects help enable our customers to automatically create critical healthcare workflows (like creating tasks) through natural language. Arjun also started development work on using LLMs to figure out open and close times for phone numbers to help solve a crucial pinpoint for our internal back office operations team.

Saskia Botha: Multilingual call support to better serve patients

During her 12-week internship, Saskia successfully led the initial phase of launching Spanish language call support for our front-office AI agents. She worked with Anish Johnson and Omar Burney to operationalize end-to-end Spanish patient-facing calls to better serve clients who have a significant Spanish-speaking patient base. Saskia coordinated across multiple teams, including Engineering, Customer Success, and Conversation Design, to implement key changes necessary for this feature. Her accomplishments include integrating a “preferred language” setting for tasks, and modifying the conversation builder and operator portal to appropriately display Spanish utterances and transcripts. This foundational work culminated in the project going live and set the stage for future phases of multilingual support, including automation and language switching.

Sachin Shankar Balasubramanyam: Leveling up Operations team training

Sachin partnered with Gavin Harding to develop a training agent to simulate operations workflows, creating configurable components that allow the leveraging of generative models for response generation in a highly deterministic way. This has allowed the Operations team to create training scenarios with increased efficiency, and dramatically decreased the maintenance costs for these scenarios

Ali Mohammadi: Scaling a privacy-safe call analysis solution

Working with Manas Pahlde, Ali worked on LingVarBench, a novel project that tackles the high cost and privacy concerns of labeling phone call transcripts. It does this by using a three-step synthetic data pipeline:

1. Generation: The system uses a large language model (LLM) to create realistic, conversational utterances that mimic a phone call.

2. Validation: A separate LLM-based extractor validates the quality of this synthetic data by checking if it can accurately retrieve the original structured information.

3. Optimization: The project then uses the validated data to automatically create and refine extraction prompts with DSPy’s SIMBA optimizer, removing the need for manual prompt engineering.

This work resulted in a significant improvement in accuracy on real customer transcripts, demonstrating that the conversational patterns learned from synthetic data can effectively generalize to real-world scenarios. Ultimately, LingVarBench shows that using synthetic, variation-rich benchmarks can match and even outperform human-tuned prompts, providing a scalable and privacy-safe solution for phone call analysis.

We’re incredibly thankful for all the hard work our team of interns contributed this year, and can’t wait to see the amazing contributions to healthcare and AI they make in the future. And of course, we’re eager to see all that our next class of interns can do. If you’re interested in learning more about working at Infinitus, check out our Careers page or talk to our AI candidate navigator – which can answer any questions about our work, our culture, our team, and more.