Skip to content

CHAT_BOT.md

Latest commit

 

History

History
99 lines (64 loc) · 7.59 KB

CHAT_BOT.md

File metadata and controls

99 lines (64 loc) · 7.59 KB
Raw

AWS Bedrock and Lex

Amazon Lex is a service for building conversational, voice-and-chat interfaces (chatbots) using Natural Language Understanding (NLU), while Amazon Bedrock is a managed service providing API access to foundation models (LLMs) for generating content and reasoning. Lex acts as the conversational "front-end," whereas Bedrock acts as the AI "brain" that powers advanced, generative responses.

Key Differences:

  • Primary Function: Lex manages intent classification, slot filling (collecting information), and dialogue management (flow). Bedrock provides access to generative models (e.g., Anthropic Claude, Amazon Titan) to answer complex queries or generate text.
  • Use Case: Use Lex for structured, transactional tasks (e.g., "book a flight" or "check account balance"). Use Bedrock for unstructured, creative, or knowledge-based tasks (e.g., summarizing documents or drafting emails).
  • Control vs. Flexibility: Lex allows for precise, deterministic control over bot conversations. Bedrock offers flexibility in choosing from multiple, powerful, but non-deterministic models.
  • Integration: Lex can use Bedrock to power its QnAIntent for retrieving answers from company data (via RAG - Retrieval-Augmented Generation).

How They Work Together: In modern applications, they are often combined, with Lex acting as the user interface and Bedrock providing intelligent, dynamic responses. A Lex bot can handle basic inputs and, when faced with a complex question, call upon Bedrock to generate an answer, thus combining structured dialogue with generative AI, say sources at Amazon Web Services.

RAG

As a software engineer already using AWS, the most efficient approach in 2026 is to build a serverless RAG (Retrieval-Augmented Generation) application using Amazon Bedrock. This bypasses the limitations of HubSpot’s built-in rule-based bots while maintaining full control over your logic and data.

Architecture Overview

  1. Frontend: A lightweight React or vanilla JS chat widget on your site.
  2. Orchestration (AWS Lambda): Acts as the brain, receiving user input, querying your data, and calling HubSpot's API.
  3. LLM & RAG (Amazon Bedrock): Uses "Knowledge Bases for Amazon Bedrock" to index your website content (stored in S3) and provide generative answers.
  4. CRM Integration (HubSpot API): Lambda uses a Private App Access Token to create contacts or send internal notifications.

Step-by-Step Implementation Guide

1. Knowledge Retrieval (RAG)

To answer questions about your site content and availability, use Knowledge Bases for Amazon Bedrock:

  • Data Source: Crawl your website or export your content to an S3 bucket as text or PDF files.
  • Vector Database: Use the "Quick Start" option in Bedrock to set up a managed vector store (like Amazon OpenSearch Serverless).
  • Availability Data: Include a specific "Availability.txt" file in your S3 bucket that you update with your current project status. The LLM will prioritize this context when asked about work.

2. Tool Use (Function Calling)

To allow the bot to "do tasks" like adding contacts, use Agents for Amazon Bedrock:

  • Define a JSON Schema for a tool called add_hubspot_contact.
  • When the LLM detects an intent to "sign up" or "contact me," it will pause and output the structured data (name, email).
  • Lambda Execution: Your Lambda function receives this structured output, calls the HubSpot Contacts API, and returns a success message to the LLM.

3. Secure HubSpot API Integration

  • Authentication: Create a Private App in your HubSpot portal to get an access_token. Store this token in AWS Secrets Manager—never hardcode it in Lambda.
  • Contact Creation: Use the POST /crm/v3/objects/contacts endpoint.
  • Internal Messaging: To notify yourself, you can use the HubSpot Notifications API or simply have the Lambda function send an email via Amazon SES.

4. Deployment via AWS Lambda URL

  • Enable a Function URL for your Lambda to provide a direct HTTPS endpoint for your website's chat widget.
  • Security: Implement a simple API key check or verify the origin of requests in your Lambda code to prevent unauthorized use of your LLM tokens.

Why this is better than HubSpot's native bot:

  • Cost: You pay only for what you use on AWS (Bedrock tokens + Lambda execution) rather than paying for a $450+/month HubSpot Professional seat.
  • Flexibility: You can switch between models (Claude 3.5, Llama 3, etc.) within Bedrock as better ones are released in 2026.
  • Customization: You can program complex logic for your "availability" questions that a standard rule-based bot cannot handle.

build-your-first-generative-ai-chatbot-with-aws-bedrock

Amazon Bedrock does not have a traditional "Always Free" tier based on usage volume. However, as of a major update in July 2025, AWS significantly revamped its Free Tier to include credit-based offers that effectively allow you to use Bedrock for free as a new customer.

1. 2026 Free Credit Offers

If you create a new AWS account, you can receive up to $200 in credits that apply to paid services like Amazon Bedrock:

  • Sign-up Bonus: New customers automatically receive $100 in credits upon account creation.
  • Builder Quests: You can earn up to another $100 by completing five onboarding activities. One of these specific tasks is often based on Amazon Bedrock (e.g., submitting a prompt in the text playground).
  • Duration: These credits are valid for 6 months from the date of account creation.

2. Bedrock Pricing Components to Watch

Even with credits, you must manage your architecture carefully to avoid high costs, particularly for the project you described:

  • Model Inference: This is billed per token (input/output). This is typically very low cost for small personal projects.
  • Knowledge Bases (Vector Store): This is the most expensive part of your proposed "answer from site content" architecture. Using Amazon OpenSearch Serverless as the vector store for RAG often has a minimum monthly cost of around $700/month.
    • Alternative: For a personal project, consider using a cheaper vector store option like Pinecone's free tier or an S3-based retrieval system to avoid this charge.
  • Agents for Bedrock: These may incur separate charges based on usage and active resource consumption during tool calls.

3. Complementary Always Free Services

While Bedrock itself is paid (via credits), the other components of your 2026 architecture are part of the AWS Always Free tier:

  • AWS Lambda: 1 million requests and 400,000 GB-seconds of compute time per month.
  • Amazon S3: 5 GB of standard storage (useful for storing your website content files).
  • Amazon DynamoDB: 25 GB of storage (useful for logging chat history or tracking availability).

Summary Table

Feature Free Tier Status (2026)
Foundation Models No "always free" usage; uses up to $200 in sign-up credits.
Prompting Playground No charge during the "Builder Quest" to earn credits.
Model Import No charge to import a custom model.
Orchestration (Lambda) Always Free within monthly limits.