Engineering blog
Architecting Intelligence
Designing a RAG system that can prove it works
From a zero-dollar Faiss + llama.cpp MVP to a $1.3M/month cloud bill at 1,000 RPS — architecture, fracture points, and stage-by-stage evaluation.
Optimizing a RAG system: five bottlenecks, two escape paths each
How a zero-dollar Faiss MVP fractures five ways — ingestion, embeddings, index, retrieval, generation — and why 94% of the $427,700/month enterprise bill is the LLM.
Should we use BM25 everywhere?
How TechPrufer built a custom BM25F-lite resume search — no LLM, no embeddings, no hardcoded skills.
Sequential thinking for agents
How structured reasoning loops improve tool use and reliability.
Cognitive architecture
Synthetic reason and the shift from probabilistic to deterministic AI workflows.
Augment context
Semantic search is replacing grep for understanding complex codebases.
India country brief
Why India ranks 2nd globally in AI usage and its productivity leap.