Caleb Delnay
Software Architect & AI Engineer
Profile
Hello there! I'm Caleb and I have 18 years experience as a technology professional across many disciplines. I move through the world with curiosity and a strong appetite to always be learning. I bring a level-headed style to software engineering with a focus on practical solutions that are both measurable and value driven to solve real world business challenges. My solutions are typically a blend of leading edge and mature technology paired with high quality engineering along with organizational cultural influence toward the future.
AI Experience
Gordon Now Chat
Gordon Now Chat is a suite of internal and external AI agents for Gordon Food Service. The AI agents are built using a variety of tools from Google Cloud Platform, including Dialogflow CX, Vertex AI, and Gemini.
I have been responsible for architecting and supervising the implementation of AI features used by the AI agents. Features were written by a team of developers with my oversight and direct contribution.
Some of the AI features I've helped create: When a potential or existing customer provides one of GFS' AI agents with an upload of their food menu (PDF or images), the agent uses a multi-step LLM process to extract likely ingredients from the menu items and then a backend search of GFS' product catalog to match the ingredients with GFS products. The matched products are used to create an order guide for the prospective or existing customer. Sentiment analysis on live chat is often used to determine when automatic escalation to a live human agent could help the end user. In the foodservice industry, words like "uncooked", "frozen" and "raw" frequently appear in product names or preparation instructions. Typical sentiment analysis models will report negative sentiment with these words and erronrously escalate to a human agent instead of providing answers. To resolve this, I supervised the creation of a custom finetuned sentiment analysis model that uses additional data to better categorize sentiment of commonly used foodservice industry terms. To provide fast lookups of customer prospects for sales people, I architected a vector search system using Vertex AI. It regularly receives a copy of the customer prospect data and makes it available for lightning fast searching by processing the data to create embeddings. Searching is surfaced to end users through natural language in our AI agents. Customers have a wide variety of questions about GFS products. I oversaw the implementation of a feature whereby our customer facing AI agent can answer product questions by looking up products by name in our product search, fetching all of the details about the product, and then passing the product details and user question into a LLM with a prompt to generate an answer. This allows for a huge variety of questions to be answered without having to manually build traditional Q&A flows in the chat design. I've helped implement a number of custom grounding solutions that allow AI agents to give accurate answers in specific knowledge domains. These solutions typically involve providing a slew of documents and data that are vectorized and then incorporated into LLM interactions to produce domain specific answers.Menu to Order Guide
Custom Sentiment Analysis for Chat
Vector Search of Customer Prospects
Questions on Product Details
Custom Grounding on Enterprise Knowledge
AI Workshops
I've been a part of multiple AI workshops in collaboration with Google that have helped to elevate the "state of the possible" to practically engineered AI solutions. Google Agentspace is a new product that acts as an agent platform for employees to easily consume and build AI agents, discover enterprise knowledge, and use LLM comprehension with enterprise data grounding. I participated in the first workshop for Agentspace and have worked closely with Google on its adoption at Gordon Food Service. Practical application of LLMs to real business challenges was the focus of a workshop with Google. Out of this workshop I began to incorporate LLMs as a core part of problem solving and building solutions at Gordon Food Service, leading to some of the AI driven features mentioned previously.Agentspace
Large Language Models (LLMs)
Work Experience
Gordon Food Service
Lead Application Architect
- Research and implement AI (ML, LLM) in customer facing applications.
- Architect customer and employee chatbots with deep authentication and authorization into enterprise data and action oriented experiences.
- Participate in transformative AI technology workshops to elevate “state of the possible” to practically engineered AI solutions.
- Lead and mentor a high-performing development team, fostering a quality focused, collaborative, and results-driven team culture.
- Set overall technical direction and engineering practices for the team.
- Partner with product team to transform and prioritize product roadmap features as a planned series of development tasks.
- Perform exploratory software engineering for continuous incorporation of state of the art technology and increased team development velocity.
- Assist with defining and reviewing application development standards.
- Network across the organization to influence technology choices, security assessments, SDLC architecture, and org best practices.
Application Architect
- Modernized customer identity and improved customer security by leading migration of legacy identity platform to Okta Cloud.
- Increased feature delivery velocity by building a custom containerized application and CI/CD platform in GCP using Kubernetes.
Web Architect
- Established development best practices for multiple Drupal sites.
- Implemented Google Analytics strategy across many sites.
Fullstack Developer, Marketing Technology
- Built and maintained corporate internet site www.gfs.com in Drupal.
Technology Specialist, Internal Communications
- Built and maintained corporate intranet site
- Built and maintained corporate digital signage