[00:00–00:45] Introduction to RAG assistant: Lena Hall introduces building and deploying a Retrieval Augmented Generation (RAG) chatbot with evidence-based Q&A capabilities. She shares a story about being stranded at an airport due to flight delays and explains how a bot can parse complex contract policies.
[00:45–01:24] System architecture overview: Explanation of the system’s architecture diagram, including the Linode Kubernetes Engine, FastAPI application container, LangChain/LangGraph orchestration, Postgres databases (pgvector and conversation state), Linode Object Storage, and OpenAI API integration.
[01:24–02:29] Technology stack rationale: Discussion on why specific tools were selected for this production build, including Python with LangChain and LangGraph for conversation state, pgvector on Postgres to avoid expensive dedicated vector databases, and Linode Kubernetes Engine (LKE) on Akamai Cloud with S3-compatible object storage.
[02:29–02:54] Phase 1: Infrastructure as code (Terraform): Setting up the infrastructure using Terraform to create the Kubernetes cluster, S3 object storage buckets, and two Postgres databases (one for vectors and one for chat history).
[02:54–03:07] Generating Linode API Key: Walkthrough of logging into the Akamai Cloud console to create a Linode Personal Access Token with read/write permissions for Databases, Linodes, Kubernetes, and Object Storage.
[03:07–03:48] Executing Terraform commands: Navigating to the terraform/ directory, copying terraform.tfvars.example to terraform.tfvars, inserting the API key, running terraform init, terraform plan, and terraform apply, followed by saving terraform output .json into terraform.output.json.
[03:48–04:12] Phase 2: Ingestion: Using s3cmd sync with credentials extracted from the Terraform output to upload local policy PDF documents to the newly created S3 object storage bucket in Akamai Cloud.
[04:12–05:16] Phase 3: Configuring the app platform: Retrieving initial admin credentials via kubectl from cluster secrets, signing into the Akamai web interface, setting up the API token for internal object storage, and installing the Harbor application to host container images.
[05:16–05:39] Managing secrets and network policies: Switching to Team View, connecting a GitHub repository, building the container image, setting up four sealed secrets (OpenAI API key, database connection strings, S3 storage credentials, and admin key), and defining outbound network policies.
[05:39–06:16] Deploying workloads and creating services: Configuring the Helm chart deployment workload (rag-chatbot) with environment variables and S3 bucket names, then creating a service to expose the workload on port 80.
[06:16–07:20] Phase 4: Initialization and validation: Testing the health endpoint via curl, initializing the Postgres database tables and pgvector extension via an admin endpoint request, triggering the document indexing background job, and interacting with the live chat interface to query Delta Airlines flight delay policies.
[07:20–08:50] Architectural bottlenecks and future upgrades: Analysis of relying on external LLM APIs (OpenAI) versus self-hosting open source models (like Llama or Mistral) on GPU-based nodes within Akamai Cloud for improved latency, control, and data sovereignty.
[08:50–09:36] Conclusion: Final remarks and invitation to “Learn AI with Lena Hall” and “Droid AI.”
[09:36–11:40] Consultation services promo: Overview of individual consultations, technical 1:1 guidance, AI integration audits, and business leader coaching options provided by Droid AI.