Implementing Exactly-Once Stateful Streaming with Apache Pulsar and RocksDB
A deep dive into combining Apache Pulsar’s messaging backbone with RocksDB’s embedded state engine to achieve exactly-once semantics in stateful stream processing.
Architecting Exactly-Once Processing with Apache Pulsar: Transactions, Fencing, and Failure Recovery
This post explores how to achieve exactly‑once semantics in Apache Pulsar by combining transactions, fencing, and failure recovery techniques for reliable streaming pipelines.
A BM25 Sparse Retriever + Tiny Transformer Generator RAG Pipeline in Pure Python
This guide walks you through implementing a BM25 sparse retriever and a tiny decoder‑only transformer generator from scratch in pure Python, producing a functional RAG pipeline you can showcase on your CV.
Build a From-Scratch LLM Sampler: Temperature, Top-k/Top-p, and Multinomial Sampling vs. Hugging Face
A hands-on guide to building a from-scratch LLM sampler with temperature scaling, top-k/top-p warping, and multinomial sampling, benchmarked against Hugging Face. Includes runnable code and a roadmap to production.
Building User-Safe Systems: Architecture Patterns That Protect Real People
User safety isn’t a feature — it’s an architectural constraint. Learn the patterns production systems use to protect real people from real harm.