Beyond Vector Search: Engineering a Two‑Stage Reranking Pipeline for Production RAG
When I upgraded the zero‑downtime RAG system that streams Stripe’s blog, the bi‑encoder kept pulling dozens of loosely related posts, and the LLM s…
Latest posts
From Code to Insight: Building a Production‑Ready Semantic Search on GitHub Repos
Ever typed a keyword into GitHub’s search bar, only to scroll past dozens of repos that *look* relevant but miss the ones that actually solve your …
Beyond Vector Search: Building a Dynamic Knowledge Graph Engine for Generative AI Pipelines
By programmatically extracting relational triplets from raw API data using structured schemas and modeling them with NetworkX, we can transform iso…
Zero‑Downtime RAG for Stripe’s Blog: A Full‑Stack MLOps Pipeline from RSS to Summarizer
Many data scientists and ML engineers, myself included, have felt the frustration of building a brilliant generative AI prototype—say, a summarizer…