ProjectsAgents
Jarvis AI
Multi-tier retrieval-augmented generation on AWS that answers questions from an internal wiki, with retrieval choices backed by benchmarks.
- Year
- 2025
- Area
- Agents
- Built with
- AWS Bedrock, OpenSearch, Lambda, SQS
Jarvis AI answers engineers’ questions from the internal wiki. The pipeline scrapes pages through SQS, embeds them with Lambda, searches with OpenSearch and answers with Bedrock.
Before choosing a design, I compared chunking strategies (fixed-size, sentence-based and semantic), dense, sparse and hybrid retrieval, and several embedding models on retrieval quality.