RAG evaluation: fix retrieval before you blame the model
Most RAG failures are retrieval failures. A practical playbook for retrieval evaluation: golden sets, recall@k, chunking tests and drift monitoring.
Production notes on AI systems, EdTech and mobile
Long-form technical writing by Lokesh Jaiswal, from twenty years of shipping production systems across EdTech, GovTech and enterprise mobile. Each piece works through what holds up after launch: evaluation, retrieval, validation, scaling and delivery.
Latest
Embedding model migration: re-indexing 10M+ vectors without downtime
A better embedding model is only an upgrade if you can install it. The sizing, dual-write and cutover plan for re-indexing 10M+ vectors with no downtime.
Layered validation for agent outputs: what each layer can actually prove
Most agents have one validator and it passes well-formed nonsense. The four validation layers, what each can prove, and how to test guardrails in CI.
Agent evaluation: building a regression harness for non-deterministic systems
Agent evaluation breaks when a flaky score reads as a regression. Mine a golden set from real traces, measure the noise floor, gate on the right thing.
Redis sharding strategy: when a single node stops being the answer
Most Redis sharding goes wrong before the first shard exists. Diagnose the real ceiling, design keys around hash slots, and reshard without a p99 spike.
Pillars
Agentic AI in production
How to run LLM agents as production software rather than as demos: evaluation harnesses, validation layers, cost control and failure handling. Written for the engineers who own an agent after launch.
Scalable AI architecture
The infrastructure underneath AI features: retrieval, vector stores, caching, sharding and latency budgets. Each piece works through the arithmetic and the failure modes before the tooling.
EdTech and assessment infrastructure
Building assessment and learning systems that hold up at institutional scale: integrity, item quality and deployment into schools. Notes from building AI assessment platforms for schools across India.
Flutter and mobile at scale
Engineering mobile apps for large device fleets and long-lived codebases: architecture, release pipelines, offline-first design, and when Flutter is the wrong choice.
Founder and operator lessons
What running several technology companies at once teaches about delivery, cash flow, procurement and where a technical founder’s time goes. Specific and practical, written for other builders.