Long-form articles on databases, cloud and DevOps, and AI, with the code included.https://naveenchandra.co.in/enhttps://naveenchandra.co.in/fine-tuning-vs-rag-vs-prompting-which-problem-belongs-to-which/https://naveenchandra.co.in/fine-tuning-vs-rag-vs-prompting-which-problem-belongs-to-which/Fine-tuning vs RAG vs prompting: RAG fixes what a model knows, prompting what it does, fine-tuning how it says it. Name the failure, then pick the fix.Fri, 26 Jun 2026 11:00:00 GMTAIcontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/jsonb-vs-separate-tables-when-normalisation-stops-paying/https://naveenchandra.co.in/jsonb-vs-separate-tables-when-normalisation-stops-paying/JSONB vs separate tables in Postgres: when a document column earns its place, what it costs the planner and your writes, and five steps to promote a key.Fri, 19 Jun 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/serverless-vs-containers-the-honest-cost-comparison/https://naveenchandra.co.in/serverless-vs-containers-the-honest-cost-comparison/Serverless vs containers on AWS: fully used Lambda compute costs about twice an equivalent Fargate task, so duty cycle decides which wins.Fri, 12 Jun 2026 11:00:00 GMTCloud & DevOpscontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/read-replicas-vs-caching-vs-both-the-decision-matrix/https://naveenchandra.co.in/read-replicas-vs-caching-vs-both-the-decision-matrix/Read replicas vs caching: a replica adds read capacity, a cache removes repeated work. A 7-row decision matrix for choosing one, both or neither.Fri, 05 Jun 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/how-to-actually-evaluate-an-llm-in-production/https://naveenchandra.co.in/how-to-actually-evaluate-an-llm-in-production/Evaluate an LLM in production with a golden set from your own traffic, a calibrated judge, a CI gate on every prompt change, and online signals.Wed, 27 May 2026 11:00:00 GMTAIcontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/zero-downtime-database-migrations-patterns-and-pitfalls/https://naveenchandra.co.in/zero-downtime-database-migrations-patterns-and-pitfalls/Zero-downtime database migrations in Postgres: lock_timeout, expand and contract, batched backfills, staged cutovers and 4 questions to pick a pattern.Thu, 21 May 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/aws-services-youre-paying-for-and-probably-dont-need/https://naveenchandra.co.in/aws-services-youre-paying-for-and-probably-dont-need/AWS services you're paying for and probably don't need: NAT Gateway traffic at $0.045 per GB, EKS cluster fees, CloudWatch log ingestion and idle resources.Sun, 10 May 2026 11:00:00 GMTCloud & DevOpscontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/indexing-strategies-when-each-index-type-actually-pays-off/https://naveenchandra.co.in/indexing-strategies-when-each-index-type-actually-pays-off/Indexing strategies for Postgres: when B-tree, GIN and BRIN (three of its six built-in index types) pay off, plus partial, covering and unused indexes.Wed, 29 Apr 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/when-to-use-each-aws-database-service/https://naveenchandra.co.in/when-to-use-each-aws-database-service/When to use each AWS database service: RDS Postgres as the default, and the measured pressure that justifies Aurora, DynamoDB, ElastiCache or OpenSearch.Thu, 16 Apr 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/database-choices-that-quietly-kill-your-aws-bill/https://naveenchandra.co.in/database-choices-that-quietly-kill-your-aws-bill/Database choices that quietly inflate your AWS bill: oversized RDS, Aurora I/O above 25% of spend, idle DynamoDB capacity, cross-AZ traffic and snapshots.Sat, 04 Apr 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/why-your-rag-system-is-actually-a-search-problem/https://naveenchandra.co.in/why-your-rag-system-is-actually-a-search-problem/Your RAG system is a search problem: hybrid search fused with reciprocal rank fusion (k = 60), cross-encoder reranking and better chunks come first.Sun, 22 Mar 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/the-real-cost-of-an-llm-call-vs-a-sql-query/https://naveenchandra.co.in/the-real-cost-of-an-llm-call-vs-a-sql-query/LLM call vs SQL query: at $2.50/$15 per 1M tokens, a 1,000-in, 500-out model call costs a cent. A query on a provisioned database has no per-request fee.Sun, 08 Mar 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/most-teams-dont-need-a-vector-database/https://naveenchandra.co.in/most-teams-dont-need-a-vector-database/Do you need a vector database? A million 1,536-dimension embeddings is about 6 GB, so pgvector in your Postgres may be enough. A 4-question check.Sat, 28 Feb 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)https://naveenchandra.co.in/postgres-does-more-than-you-think/https://naveenchandra.co.in/postgres-does-more-than-you-think/Postgres does more than you think: full-text search built in since 2008, JSONB, pgvector and SKIP LOCKED queues, and when a dedicated system earns its place.Fri, 20 Feb 2026 11:00:00 GMTDatabasescontact@naveenchandra.co.in (Naveen Chandra)