Ad Astra AI

A technical hub for production-ready AI engineering projects. Learn best practices for building, deploying, and scaling AI systems in real-world production environments.

📬Subscribe on Substack

Featured Projects

RAG

2025-01

25 min

SentiWiki AI: End-to-End Agentic RAG for ESA Copernicus Sentinel Missions

Production-grade agentic RAG system deployed on AWS ECS for querying ESA Copernicus Sentinel mission documentation with 96.1% faithfulness.

LangGraphQdrantFastAPIAWS ECSNext.jsDocker

SentiWiki AI is an end-to-end agentic RAG system that treats documentation retrieval as an intelligent engineering problem. It distinguishes between looking up a fact (Retrieval) and deriving an answer (Reasoning), using LangGraph-based agents to route queries, decompose complex questions, and retrieve grounded answers from 66+ SentiWiki documents. The system is built as a production-grade, fully deployed stack on AWS ECS Fargate, demonstrating real-world engineering practices including end-to-end data pipelines, agentic intelligence, hybrid retrieval, asynchronous FastAPI backend, and complete observability.

LangGraphQdrantFastAPILiteLLMNext.jsRAGASAWS ECSPrometheusGrafanaClaude 3Crawl4AIDockerRAGProduction SystemsSentinel Missions

View on GitHub 📬Read on Substack

Learning Path (Blog Series)

View All on Substack

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 1): Project Overview

25 min

Welcome to Ad Astra AI — a technical hub where I'll be building production-ready AI engineering projects. This is the entry point to a full series that explains every part of the SentiWiki AI system, from data scraping to AWS deployment.

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 2): Data Scraping with Crawl4AI

20 min

How SentiWiki is scraped, cleaned, and turned into markdown ready to be ingested. Learn about Crawl4AI, Breadth-First Search strategies, and cleaning pipelines for RAG-ready data.

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 3): Chunking & Ingestion

18 min

Sentiwiki structure-aware chunking, metadata, and local embeddings, and ingestion.

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 4): Advanced Retrieval Pipeline

Coming Soon!

Hybrid search (semantic + keyword), Qdrant, reranking, and retrieval strategies.

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 5): Agentic RAG with LangGraph

Coming Soon!

LangGraph agents for routing, query rewriting, grading, and multi-step flows.

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 6): Evaluation & Monitoring with RAGAS, adversarial test, Prometheus/Grafana and LangSmith

Coming Soon!

RAGAS evaluation, adversarial test sets, metrics, Prometheus/Grafana dashboards, and LangSmith traces.

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 7): Backend with FastAPI & Frontend with Next.JS

Coming Soon!

FastAPI API design, real streaming, Dockerization, and the Next.js chat UI.

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 8): Docker Compose & CI/CD with GitHub Actions

Coming Soon!

Docker Compose setup and automated CI/CD pipelines with GitHub Actions.

Building an end-to-end agentic RAG deployed on AWS for ESA SentiWiki (Part 9): AWS Deployment with ECS/Fargate architecture, ALB, Secrets Manager, and S3

Coming Soon!

ECS/Fargate architecture, Application Load Balancer, Secrets Manager, Cloud Map, scaling and costs.

Stay Updated with Ad Astra AI

Get weekly deep-dives on production-ready AI engineering projects directly to your inbox. Learn best practices for building, deploying, and scaling AI systems in real-world environments.

📬Subscribe on Substack