Vericant, an educational testing company, developed and deployed an AI-powered video interview analysis system in just 30 days. The solution automatically processes 15-minute admission interview videos to generate summaries, key points, and topic analyses, enabling admissions teams to review interviews in 20-30 seconds instead of watching full recordings. The implementation was achieved through iterative prompt engineering and a systematic evaluation framework, without requiring significant engineering resources or programming expertise.
This case study documents how Vericant, a third-party video interview company serving high schools and universities with their admissions processes, rapidly deployed a generative AI solution to help admissions teams process interview content more efficiently. The company was later acquired by ETS (Educational Testing Service), the world’s largest private nonprofit educational testing organization. The discussion is presented as a conversation between Kevin (the interviewer/host) and Guy Savon, co-founder and CEO of Vericant, recorded approximately one month after they had an initial conversation about potential generative AI use cases.
Vericant’s core business involves making interviews scalable for educational institutions. They conduct video interviews (approximately 15 minutes each) with applicants, particularly international students, to help admissions teams get a more holistic view of candidates and assess their English speaking capabilities. The fundamental challenge they addressed was that while Vericant had successfully scaled the interview collection process (handling tens of thousands of interviews), they hadn’t solved the problem of helping admissions teams actually process and consume all those interviews efficiently.
Admissions officers faced a significant time bottleneck: to understand what happened in each 15-minute interview, they had to watch the entire video. This didn’t scale well when dealing with large numbers of applicants. The company recognized an opportunity to use generative AI to dramatically reduce this processing time while still providing meaningful insights from each interview.
The most notable aspect of this case study is the low-code, rapid prototyping approach. Guy Savon, the CEO, built the initial solution himself without any programming. The key tools used were:
OpenAI Playground: This was the primary development environment. The playground allowed direct interaction with GPT models using a system message (containing the prompt) and user message (containing the interview transcript). Guy would paste prompts into the system message section, paste interview transcripts into the user message section, and observe the outputs.
Google Sheets: Used as both a data management tool and evaluation framework. Outputs from the OpenAI playground were copied back to spreadsheets for review and scoring.
The system prompt evolved through iterative refinement. The initial prompt structure included:
The prompts were refined based on observing actual outputs and identifying patterns of errors or suboptimal results. This is a practical example of prompt engineering in action, where the CEO personally iterated on the instructions until the outputs met quality standards.
A particularly noteworthy aspect of this implementation is the systematic evaluation framework. Rather than relying on subjective impressions, the team implemented a structured approach:
This evaluation framework was inspired by content about dealing with LLM hallucinations, specifically a video about creating evaluation frameworks to determine whether outputs are usable for specific use cases.
The case study clearly demonstrates an iterative refinement cycle:
Once the prompts were sufficiently refined, they were handed off to the engineering team for integration. The production feature, called “AI Insights,” was integrated into the existing admissions portal where teams access interview content. The feature includes:
The naming decision is interesting from a product perspective—they deliberately chose to include “AI” in the feature name (AI Insights) rather than something like “Vericant Assistant.” The reasoning was that at the current moment in AI adoption, there’s more user forgiveness for AI-generated content, and being transparent about the AI nature of the feature sets appropriate expectations.
The deployment also included appropriate disclaimers indicating the feature was in beta and requesting feedback, demonstrating responsible AI deployment practices.
One of the most compelling aspects of this case study is the minimal resource investment:
This demonstrates that for certain use cases, generative AI features can be prototyped and deployed with remarkably low overhead, particularly when the technical approach (prompt engineering vs. custom model development) matches the problem complexity.
Vericant achieved first-to-market status in their specific niche. The case study emphasizes this as a significant opportunity during periods of transformative technology—being early provides substantial leverage because there are no established standards or expectations to meet.
Perhaps most significantly, this small initial project catalyzed a broader strategic vision. Before this implementation, Vericant’s value proposition was focused on scaling interviews (collecting tens of thousands of interviews). After deploying the AI insights feature, the company articulated a new complementary value proposition: scaling the admissions team’s ability to process interviews.
This represents a meaningful expansion of their business model, transforming from primarily an interview collection platform to a full-cycle interview intelligence platform. The CEO explicitly noted that this broader vision only crystallized after doing the hands-on work with the actual technology.
The case study addresses the psychological barriers to adopting new technology. When approaching something never done before, time estimates have very wide ranges (could be 2 hours or 3 weeks), creating anxiety about commitment. Having a framework and methodology reduced uncertainty and increased confidence to “just jump in.”
The CEO emphasized the value of hands-on experimentation over extended planning phases. Playing with the tools made abstract possibilities concrete, and the strategic vision emerged from practical experimentation rather than preceding it.
The case demonstrates that generative AI deployment doesn’t necessarily require specialized AI/ML expertise or significant engineering resources for initial prototyping. Using existing tools like OpenAI’s playground and Google Sheets, a non-technical CEO was able to develop and validate a production-ready prompt system.
While this case study presents an encouraging picture of rapid AI deployment, several caveats are worth noting:
Nevertheless, this case study provides a practical template for organizations considering similar applications, particularly in the education sector where processing large volumes of qualitative content (interviews, essays, applications) is common.
Snorkel developed a specialized benchmark dataset for evaluating AI agents in insurance underwriting, leveraging their expert network of Chartered Property and Casualty Underwriters (CPCUs). The benchmark simulates an AI copilot that assists junior underwriters by reasoning over proprietary knowledge, using multiple tools including databases and underwriting guidelines, and engaging in multi-turn conversations. The evaluation revealed significant performance variations across frontier models (single digits to ~80% accuracy), with notable error modes including tool use failures (36% of conversations) and hallucinations from pretrained domain knowledge, particularly from OpenAI models which hallucinated non-existent insurance products 15-45% of the time.
LinkedIn evolved from simple GPT-based collaborative articles to sophisticated AI coaches and finally to production-ready agents, culminating in their Hiring Assistant product announced in October 2025. The company faced the challenge of moving from conversational assistants with prompt chains to task automation using agent-based architectures that could handle high-scale candidate evaluation while maintaining quality and enabling rapid iteration. They built a comprehensive agent platform with modular sub-agent architecture, centralized prompt management, LLM inference abstraction, messaging-based orchestration for resilience, and a skill registry for dynamic tool discovery. The solution enabled parallel development of agent components, independent quality evaluation, and the ability to serve both enterprise recruiters and SMB customers with variations of the same underlying platform, processing thousands of candidate evaluations at scale while maintaining the flexibility to iterate on product design.
Rippling, an enterprise platform providing HR, payroll, IT, and finance solutions, has evolved its AI strategy from simple content summarization to building complex production agents that assist administrators and employees across their entire platform. Led by Anker, their head of AI, the company has developed agents that handle payroll troubleshooting, sales briefing automation, interview transcript summarization, and talent performance calibration. They've transitioned from deterministic workflow-based approaches to more flexible deep agent paradigms, leveraging LangChain and LangSmith for development and tracing. The company maintains a dual focus: embedding AI capabilities within their product for customers running businesses on their platform, and deploying AI internally to increase productivity across all teams. Early results show promise in handling complex, context-dependent queries that traditional rule-based systems couldn't address.
