SShortSingh.
Back to feed

Hestia App Uses Strands AI Agents to Flag Household Money Leaks Without Overstepping

0
·3 views

Hestia is a household finance app built for the AWS Agents for Humans hackathon that uses two Strands AI agents to help families catch overlooked expenses like lapsed warranties, unwanted subscriptions, and missing receipts. The app runs on Amazon Bedrock using Claude Haiku 4.5 and is designed so that the AI only reads and flags issues, while all date and amount calculations are handled by plain Python code. A review agent scans household data through four read-only tools and produces a short briefing, while a separate reading agent converts pasted text into proposed records — neither agent can write, send, or modify any data. To prevent the model from overstating conclusions, the system prompt explicitly bars it from implying refund entitlements or inventing deadlines, restricting it to facts drawn directly from tool outputs. The code also independently verifies the model's responses after each run, treating the prompt as a guideline rather than a guaranteed constraint.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingHacker News ·

Analysis Claims Claude AI Tends Toward Contrarian Responses

A Medium article by user rdsubhas argues that Anthropic's Claude AI model exhibits contrarian behavior in its responses. The piece was shared on Hacker News, where it gathered modest engagement with 8 points and 1 comment. The author suggests Claude has a tendency to push back or disagree rather than align with user inputs. The article raises questions about how Claude's training may influence its conversational style. The post has sparked limited but notable discussion in the developer community.