Blog
Thoughts on DevOps, database engineering, and the art of automation.
MFS Production Conversation: "You Know One Duplicate Request Could Create Financial Mismatch, Right?"
Why duplicate cash-out and send-money requests from timeouts and retries can double-debit wallets and break ledger reconciliation in an MFS system — and how idempotency keys, request tracking, and ledger-based posting prevent it.
PostgreSQL DBA Conversation: "You Know That One Big DELETE Could Hurt Production, Right?"
A senior PostgreSQL DBA's warning on why a single bulk DELETE can trigger WAL growth, replication lag, dead tuples, and locking issues — plus the batch-delete, partitioning, and retention strategies that keep production safe.
Simplifying Git Bash Prompt on Windows for a Cleaner Developer Workflow
Default Git Bash prompts display the full path and clutter your terminal. Here's how to configure PS1 and PROMPT_DIRTRIM for a cleaner developer experience.
What Is RAG? How Retrieval-Augmented Generation Works Internally
RAG is more than a vector database with an LLM. It's a structured architecture where an LLM retrieves external knowledge at runtime to generate grounded, accurate answers.
Implementing Role and Group-Based Security with Qdrant in Enterprise AI Systems
Qdrant is a vector search engine, not an enterprise IAM system. Learn the correct architecture for implementing role and group-based access control in production RAG pipelines.
New Attack Surfaces in AI Systems: Understanding the Security Risks Unique to LLM Applications
AI systems introduce attack vectors that traditional security tools don't catch. Prompt injection, indirect injection, and tool misuse are now first-class security concerns.
Why Security Matters in AI-Based Systems: Building Secure, Monitorable Enterprise AI Platforms
Many teams prioritize model quality over security architecture — a critical mistake. An AI system can unintentionally expose massive amounts of sensitive data in a single response.
How Data Is Stored and Retrieved from a Vector Database
Vector databases don't search for exact embedding values — they perform nearest-neighbor similarity searches. Here's the complete ingestion and retrieval flow explained.
Scaling HDFS: Fixing the "Protocol Message Too Large" Block Report Error
When DataNodes send Full Block Reports exceeding 64 MB, the NameNode rejects them. Here's how to fix the ipc.maximum.data.length limit in large Hadoop clusters.