Technology
We work with every aspect of architecting large language model applications. Most of the elements below are built right into the gepeto SDK, and we use best-in-class tools for 3rd party partners.
Below are our primary building blocks.
Core AI Technologies
Large Language Models
We utilize OpenAI, Anthropic, Bedrock, Llama, Groq, and Ollama as our LLM providers.
Fine-Tuning
Fine-tuning involves adapting a pre-trained LLM to perform specific tasks by training it on a smaller, task-specific dataset. For most projects, fine-tuning is not necessary.
Prompting
Crafting effective system prompts guides the model’s output more accurately. We use our own no-code prompting tool in our AI applications to iterate until we have consistent outputs.
Retrieval-Augmented Generation (RAG)
RAG combines LLMs with a retrieval system to access a larger set of examples to model responses after. We leverage RAG to enable our AI applications to provide up-to-date, context-aware responses.
AI Agents and Automation
Single Agents
We’ve developed our own agent management platform to create AI agents from scratch. They can automate complex tasks, interact with users, and handle dynamic scenarios.
Function Calling
Our AI agents can execute code or scripts as part of their operations. These function calls allow our agents to interact with other software, such as creating a new user or updating a CRM.
Agent Teams
Complex tasks require a team of agents to ensure consistent performance. Each agent handles a specific part of the process, passing information along to the next. Our platform allows us to network as many agents as needed to deliver results.
Web Crawling
Our AI agents can navigate the internet to collect and index website information. These agents autonomously gather data for purposes like market analysis, contextual awareness, or training datasets.
Multi-Channel
We can deploy agents across communication channels such as SMS, web chat, Voice over IP (VoIP), email, and or embedded voice. Essentially, any medium where humans can send and receive communication can be integrated with our AI agents.
Development and Deployment Practices
Playground
We use a no-code interface to rapidly gather feedback during testing. Users can thumb up/down messages and add notes to improve performance.
Evaluations
We continuously evaluate our AI models to ensure they perform as intended. Pre-production evaluations happen before deployment, testing capabilities and identifying issues. Runtime evaluations occur while the model is in operation.
Self-Improvement
We build our systems with self-improvement capabilities. Our applications learn and enhance their performance over time.
Observability
We build in observability from the start. This allows us to monitor our applications continuously for performance and behavior. This includes tracking metrics, logging events, and detecting anomalies.
Integrations
We created integrations with third-party clients, connecting our AI applications with external systems, services, or software. Our AI solutions work seamlessly within your existing tools.
Deployment
We manage deployment by establishing Continuous Integration and Continuous Deployment (CI/CD) pipelines. Our SDK helps set up the necessary infrastructure for scalability and security.
Ready to empower your organization with AI?
Fill out the form and our team will get in touch.