AI Agent
AI DevOps & Infrastructure
Document Processing
RAG & Embedding
Mainstay: Scalable context-aware AI Agent for real estate
“Tech 42 was a strong partner in helping Mainstay advance our document intelligence platform by building a conversational AI solution on AWS to unlock deeper insights across our CC&R and HOA document repository.
The team designed and implemented a Retrieval-Augmented Generation (RAG) agent leveraging AWS Bedrock Knowledge Bases and OpenSearch, enabling our analysts to ask natural language questions and receive accurate, source-cited responses across large document sets. In addition, they extended our existing document processing workflows to support additional HOA document types and improve prompt performance and reliability.
Tech 42 also delivered a scalable API layer via ECS and FastAPI, integrated agent memory and guardrails, and implemented monitoring through Langfuse for traceability and evaluation. The solution was delivered with a strong infrastructure foundation using Terraform, along with clear documentation, architecture diagrams, and a recorded knowledge transfer session that equipped our team to manage and deploy the system independently.
From a collaboration standpoint, the engagement ran smoothly — communication via Slack and regular check-ins were effective, and having visibility into the codebase through GitHub allowed our team to review progress proactively.
Overall, Tech 42 delivered high-quality work aligned with the scope and objectives, and we’re well positioned to take the solution forward into production. We appreciate the partnership and look forward to potential future phases together.”
Rodrigo Reis
Engineering Leader
Project summary
Mainstay needed a way for analysts to query large repositories of CC&R and HOA documents without manual review or custom data pulls. Tech 42 built a Retrieval-Augmented Generation (RAG) agent using AWS Bedrock Knowledge Bases and an OpenSearch vector database. Analysts ask natural language questions and receive accurate, source-cited responses across large document sets. The agent runs on a LangGraph framework with custom prompt engineering and Bedrock Guardrails for content safety. A FastAPI REST API deployed on ECS handles streaming and standard responses, with auto-scaling based on CPU and memory metrics. Langfuse provides monitoring and tracing for performance evaluation. Existing document processing workflows were extended to support additional HOA document types. All infrastructure is defined and managed with Terraform.
