Explore AI-powered solutions from Telnyx's Open AprIl challenge that showcase our employees’ creativity and innovation.
At Telnyx, we believe in the power of purposeful innovation, ownership, transparency, and accessibility. These core values were on full display during our Open AprIl challenge.
Open AprIl is a competition designed to inspire our team to explore and integrate artificial intelligence (AI) technology into our operations to enhance efficiency, productivity, and service quality.
The Open AprIl challenge is an aptly-named nod to the month of April—which symbolizes growth and renewal—and to . The challenge was a call to action for our talented employees. We’re proud to say they rose to the occasion, creating a variety of AI-powered solutions that have the potential to revolutionize our operations and service delivery.
Related articles
James Whedbee, Telnyx’s VP of Engineering, spearheaded the challenge. “We see vast potential for AI applications within Telnyx. We wanted to create a fun, engaging space for our team to experiment with AI and feel empowered to build their own tools,” said Whedbee.
Team members from all levels of the company met the challenge with enthusiasm and support—and the generous cash prize certainly added an extra layer of excitement.
Employees were given 32 hours, spread over the month of April, to work on their projects. The result? A plethora of innovative solutions, crafted by both solo participants and self-organized teams.
The competition was fierce, with 16 of the 50 initial entries making it to the final round. During the final round, teams created a demo proof of concept. Telnyx employees then voted on their favorite projects.
Overall, we were impressed by our teammates’ abilities to ideate and create innovative and impactful AI solutions. To honor and recognize our team members’ hard work, we’ll take a closer look at the top three projects that received the most votes in the Open AprIl competition.
We awarded the grand prize to a team that developed a customer-facing chatbot for the Telnyx website. This sophisticated tool leverages Telnyx documentation to answer customer queries efficiently.
The idea for the Customer Support Chatbot emerged during a 2022 planning session in which the Network Operation Center (NOC) Engineering team allocated time to enhance their skills and explore emerging technologies. Among several possibilities, AI and ML (machine learning) captured their interest, which sparked the idea of creating an AI chatbot that could interact with Telnyx customers.
The team noticed that, as the Telnyx customer base grew, their acknowledgment times remained the same, but their update times began to increase. They tossed around the idea of building an internal tool agents could use to verify customer questions. The goal was to reduce update times by having agents use this tool instead of requiring them to confirm information with technical managers manually.
Project overview
Telnyx NOC Engineering team members Ciaran and Dillin were the brains behind this ambitious project. Ciaran drove technical implementation of the chatbot. His extensive coding knowledge in front- and back-end development was a major factor in creating a minimum viable product (MVP) tool.
The NOC Engineering and Self-Service teams helped Dillin and Ciaran establish a standardized structure for the documentation used by the chatbot to ensure consistency and clarity across materials. The chatbot uses vector embeddings to find relevant documentation and employs GPT-4 to generate accurate, helpful responses. Lisa, a front-end developer, was instrumental in developing the widget that customers will interact with on their accounts.
Like any creative endeavor, this project had its share of highs and lows. One highlight was introducing the speed dial component, which combined the features of the newly-created support chatbot and ChatGPT for easy access to agent assistance when needed.
The team found it exciting to experiment with AI and ML, particularly since the development of AI has accelerated immensely since the start of the year. Ciaran noted that there were times when he was “writing methods that were outdated within a week because a new development had been made.”
“I enjoyed working with such an emerging and powerful piece of technology. It makes me excited to think about how much better it will get in the future and how our bot can benefit from additional technological advances,” he added.
However, there were also challenges. Ciaran mentioned the complexity of designing prompts for a large language model (LLM), a task known as prompt engineering. They had to continually tweak the wording and retest queries to ensure the bot would provide the expected output.
Project impact
The Customer Support Chatbot significantly enhances customer support efficiency by quickly and accurately addressing customer questions. It optimizes resource use by breaking down articles into bite-sized pieces for quick, to-the-point responses. It also ensures customers receive the type of comprehensive, relevant answers you’d expect from a seasoned support agent.
If the bot cannot answer a question, it will connect customers with a human support agent. However, the chatbot has access to an extensive repository of documentation, so it effectively eliminates a large volume of queries the customer support team previously had to address manually.
The result? The team can provide focused, targeted support for customers who need to interact with a human agent, and customers with straightforward questions get assistance right away.
Future applications
Looking ahead, the team has ambitious plans for the chatbot. They intend to leverage more of their internal knowledge about products to build customized models, thereby enhancing the chatbot's understanding of the telecommunications industry and the intricacies of Telnyx products.
In doing so, the chatbot will be able to provide even more technical and tailored support to our customers. Additionally, the team will continue to gather customer feedback to refine the chatbot's responses and performance. It can even offer suggestions for continuous improvement of support documentation.
We’re excited to leverage this tool to achieve even higher levels of customer support at Telnyx.
On June 26, the AI Assistant was released in beta to a subset of Telnyx users. The team plans to roll out the AI Assistant to all users in upcoming weeks. Click here to learn more about the AI Assistant, which is accessible to select users now in the Mission Control Portal.
The Incident Co-Pilot, a specialized chatbot for managing incidents, clinched the second-place spot. Enzo and Opeyemi, Senior Software Engineers, designed this chatbot to expedite incident mitigation.
This team noticed that incidents—defined, in this case, as events that impact customers—have many different root causes. However, the mitigation strategy is often the same from incident to incident. Effective communication facilitates the gathering and sharing of essential information, and system knowledge assists in making the right decisions for quick incident resolution.
Project overview
The team's project, Incident Co-Pilot, aims to enhance communication during an incident, speeding up incident resolution times. It captures all Slack messages and system knowledge related to a specific incident and consolidates the data in a central location accessible to all stakeholders. This immediate summary provides real-time updates and serves as a resource for future reference.
Incident Co-Pilot can provide instructions and incident summaries, search for past incidents based on specific criteria, and identify similar past incidents. Furthermore, the team leveraged past incident data to facilitate the resolution of current and future ones. By analyzing similar past incidents, the chatbot could handle incidents with precision and speed for proactive incident management and continuous learning.
Project impact
By providing immediate access to past incidents and instructions for various processes, Incident Co-Pilot enhances the ability to mitigate incidents swiftly and efficiently. It aids in incident summary creation and has the potential to identify similar past incidents, providing crucial insights for incident management.
Enzo believes that “integrating Incident Co-Pilot into the incident management pipeline will dramatically reduce the average resolution time at Telnyx.”
Future applications
Enzo and Opeyemi chose to focus on a breadth of features rather than a ready-to-use model in their initial demo. Their next step is to fully realize the project so it can oversee the entire lifecycle of an incident, from its creation to the completion of post-mortem actions.
Despite having no programming experience, Katie, our Senior Manager of Technical Recruiting, and Alejandra, our Senior People Operations Manager, built a chatbot using GPT-3.5 Turbo. Their chatbot is designed to address typical recruitment, HR, and onboarding queries.
Katie and Ale often found themselves swamped with repetitive questions about the interview and onboarding process. Hoping to streamline their responses and reduce delays, they decided to build a recruiting chatbot that would handle these frequently asked questions. The chatbot was born out of the desire to save the HR team hours of time each week and create a more streamlined, efficient experience for candidates.
Project overview
The chatbot improves the candidate experience by providing 24/7 support for common questions related to the interview process and onboarding. By leveraging a question-answer database, it answers inquiries efficiently, saving time for the HR team and providing timely responses to prospective candidates. It streamlines team workflow by handling common questions. Instead of devoting time to easy-to-answer questions, Katie and Ale can focus on in-depth conversations with potential candidates.
The duo took on roles that suited their backgrounds. While neither had prior coding experience, Katie tackled the coding challenge by leveraging ChatGPT and Replit to build the project in Python. Alejandra, on the other hand, focused on compiling the FAQ file that would fuel the chatbot's responses.
“Seeing my code actually run and not break was pretty amazing,” Katie said. “I loved the learning process, and I feel like I’ve gained so much from this experience. I most definitely will continue coding in my spare time!”
Project impact
This chatbot could significantly improve the candidate experience by providing round-the-clock support for common recruitment queries. It leverages a question-answer database to respond to inquiries efficiently, thus saving time for the HR team and providing immediate responses to prospective candidates.
Future applications
Katie and Alejandra are looking forward to getting the chatbot up and running. They already have plans to enhance the bot's functionality. Their first step is to integrate the frontend and backend of the chatbot and expand its knowledge base to include detailed interview guides for all roles.
Katie mentioned taking inspiration from Ciaran and Dillin’s chatbot project. She added that she’d like to “use embeddings and machine learning so the bot can answer the questions more naturally and accurately” in future iterations.
While our winners certainly made a remarkable impression, the competition was fierce, and many other noteworthy projects also deserve recognition. In this section, we’ll highlight five other entries that brought considerable potential to the table with their cutting-edge ideas and commendable dedication.
Working collaboratively, Niamh, Ruchita, and Nate from the SalesOps team partnered with Kevin in Business Development to devise an innovative AI Prospecting tool. Drawing on the capabilities of ChatGPT, Slack, Zapier, and Salesforce, AI Prospector offers suggestions for potential businesses that mirror recently converted clients. The goal is to empower sales representatives by providing a framework to duplicate their successful selling methodologies.
What’s next: Integrating Salesforce functionality for better tracking and potentially using the ZoomInfo API to ensure suggested leads are not existing customers or already part of the sales funnel.
Hubert, Marcin, and Michal from the TelApps team—the team responsible for building and maintaining the software that allows Telnyx customers to programmatically initiate and control phone, fax, and video calls—created Call Control Summaries. Call Control Summaries is tool that could potentially use ChatGPT to extract, summarize, and present synopses of phone conversations started with Telnyx's Call Control API. The tool is also capable of carrying out sentiment analysis.
What’s next: Further integrating this tool within Telnyx's systems with potential improvements, including a Call Control API integration and more detailed sentiment analysis.
Leszek, a Senior Solutions Engineer, built a Pricing Tool Automation project that uses GPT-4 to automate the creation of offers for outbound voice using different channels like text chat, voice bot, and SMS.
What’s next: Further optimizing the tool to handle more complex pricing models and automate more aspects of the offers created by AEs.
Segun and Colton from the Finance team developed Vendor Investigations, a chatbot that interacts with their transactional database. This tool helps investigate transaction data and spot trends and patterns at the vendor level.
What’s next: Further integrating the bot into Slack for easy accessibility and inquiries.
Dylan, Jake, Jonathan, and Phillip of the Messaging Operations team developed the 10DLC AI Vetting and Admin Tool, an AI system that streamlines the 10DLC campaign vetting and approval process.
What’s next: Further refining the tool and potentially expanding its capabilities to handle additional administrative tasks.
The Open AprIl Contest showcased the immense talent and creativity present within Telnyx and gave us a glimpse into the immense
potential of AI. While all of these projects have made significant strides in harnessing AI for operational efficiency, we know they represent just a fraction of the innovation that took place during the Open AprIl Contest. As we move forward, we're excited to see how these projects, and others like them, continue to evolve and shape the future of Telnyx.
Stay tuned for more updates as we continue to explore the integration of these AI solutions into our operations and products. The potential for improved efficiency, customer service, and overall productivity is extensive, and we can't wait to see how these ideas develop in the coming months.
What is AI agent evaluation and why does it matter? AI agent evaluation is the systematic testing of an agent’s behavior, accuracy, latency, safety, and business outcomes before deployment. Strong evaluation reduces risk, speeds approvals, and ensures reliable performance at scale.
Which metrics determine if a voice AI agent is production ready? Prioritize round-trip latency, turn-taking responsiveness, speech recognition accuracy, tool-call success rate, and containment rate. If the agent triggers post-call messages, tie outcomes to conversion and engagement metrics that reflect MMS marketing performance.
How should I build a representative evaluation dataset? Include real and synthetic voice interactions that cover core intents, accents, languages, background noise, and edge cases. If your follow-ups include media, define how MMS messaging is expected to render so reviewers apply consistent criteria.
What does an approval workflow look like for AI agents? Use a gated pipeline that moves from sandbox to pilot to general availability, with required human-in-the-loop signoff at each stage. Tie promotion to hard thresholds and regression tests, and block rollout on any safety or compliance regression.
How do I evaluate multichannel follow-ups like sending media after a call? When the agent sends a confirmation or demo via messaging, decide whether the content fits SMS or MMS to keep scenarios faithful to real delivery rules. Validate link handling, image rendering, and opt-out flows as part of the test outcomes.
What rollout and monitoring practices should follow approval? Start with canary traffic and A/B tests, then monitor latency, escalation rate, abandonment, and per-intent success in near real time. Keep a rollback plan and run continuous regression suites on every model, prompt, or tool change.
How do I gather reviewer feedback at scale without biasing results? Distribute transcripts and audio to blinded reviewers using a rubric that measures task completion, tone, and policy adherence. If you coordinate via messaging, choose MMS group or broadcast messaging based on whether you want a single shared thread or one-to-many distribution.
