In-depth technical writing on AI systems, software architecture, Laravel, and rescuing software that is in trouble.https://vishal.life/en© 2026 Vishal Rajpurohithttps://vishal.life/blog/transformer-inference-from-request-to-token/https://vishal.life/blog/transformer-inference-from-request-to-token/A systems-level tour of prefill, decode, batching, memory movement, and the scheduling decisions that determine real-world LLM latency.Fri, 24 Jul 2026 00:00:00 GMTInference SystemstransformersinferenceGPUservingVishal Rajpurohithttps://vishal.life/blog/kv-cache-engineering-paged-attention-prefix-reuse/https://vishal.life/blog/kv-cache-engineering-paged-attention-prefix-reuse/How to size, allocate, reuse, compress, and evict the state that makes autoregressive decoding practical.Sat, 18 Jul 2026 00:00:00 GMTInference SystemsKV cachepaged attentionmemorylatencyVishal Rajpurohithttps://vishal.life/blog/semantic-caching-with-correctness-boundaries/https://vishal.life/blog/semantic-caching-with-correctness-boundaries/A production design for similarity keys, freshness, authorization, invalidation, and the economics that determine when semantic reuse is safe.Sun, 12 Jul 2026 00:00:00 GMTInference Systemssemantic cacheembeddingslatencycorrectnessVishal Rajpurohithttps://vishal.life/blog/speculative-decoding-production-guide/https://vishal.life/blog/speculative-decoding-production-guide/Draft models, acceptance math, tree proposals, and the operational details that decide whether speculation makes serving faster.Fri, 10 Jul 2026 00:00:00 GMTInference Systemsspeculative decodinglatencydraft modelssamplingVishal Rajpurohithttps://vishal.life/blog/synthetic-data-for-evaluations/https://vishal.life/blog/synthetic-data-for-evaluations/How to generate, filter, diversify, and maintain synthetic cases without turning an evaluation suite into a mirror of its generator.Thu, 09 Jul 2026 00:00:00 GMTEvaluation and Securitysynthetic dataevaluationcoverageLLM judgesVishal Rajpurohithttps://vishal.life/blog/llm-gateway-as-policy-and-control-plane/https://vishal.life/blog/llm-gateway-as-policy-and-control-plane/Designing the routing, budgets, identity, resilience, and evidence layer between products and a changing model portfolio.Mon, 06 Jul 2026 00:00:00 GMTInference SystemsLLM gatewaymodel routinggovernanceresilienceVishal Rajpurohithttps://vishal.life/blog/privacy-preserving-production-ai/https://vishal.life/blog/privacy-preserving-production-ai/Purpose limitation, minimization, isolation, retention, redaction, and verifiable deletion across retrieval, models, tools, traces, and feedback loops.Fri, 03 Jul 2026 00:00:00 GMTEvaluation and Securityprivacydata governancePIIsecurityVishal Rajpurohithttps://vishal.life/blog/incident-response-for-ai-systems/https://vishal.life/blog/incident-response-for-ai-systems/A response playbook for quality regressions, prompt attacks, retrieval contamination, runaway agents, cost spikes, and provider failures.Tue, 30 Jun 2026 00:00:00 GMTProduct Architectureincident responseobservabilityagentsLLMOpsVishal Rajpurohithttps://vishal.life/blog/production-rag-evidence-pipeline/https://vishal.life/blog/production-rag-evidence-pipeline/Designing retrieval-augmented generation around ingestion quality, query planning, evidence assembly, and verifiable answers.Mon, 29 Jun 2026 00:00:00 GMTRetrievalRAGretrievalgroundingknowledge systemsVishal Rajpurohithttps://vishal.life/blog/hybrid-search-rank-fusion-engineering/https://vishal.life/blog/hybrid-search-rank-fusion-engineering/A practical design for candidate generation, score normalization, rank fusion, metadata filters, and retrieval evaluation.Sun, 21 Jun 2026 00:00:00 GMTRetrievalhybrid searchBM25vector searchrankingVishal Rajpurohithttps://vishal.life/blog/rerankers-the-quality-layer-after-retrieval/https://vishal.life/blog/rerankers-the-quality-layer-after-retrieval/How cross-encoders, late interaction, listwise ranking, and calibration turn noisy candidates into usable evidence.Fri, 12 Jun 2026 00:00:00 GMTRetrievalrerankingcross-encodersretrievalrelevanceVishal Rajpurohithttps://vishal.life/blog/agent-tool-loop-engineering/https://vishal.life/blog/agent-tool-loop-engineering/A concrete architecture for tool selection, state transitions, permissions, budgets, and recovery in production agents.Fri, 05 Jun 2026 00:00:00 GMTAgentsagentstool useorchestrationstate machinesVishal Rajpurohithttps://vishal.life/blog/durable-execution-for-long-running-ai-agents/https://vishal.life/blog/durable-execution-for-long-running-ai-agents/Persisting checkpoints, replaying deterministically, handling human approval, and surviving crashes without duplicating side effects.Wed, 27 May 2026 00:00:00 GMTAgentsdurable executionworkflowsagentsreliabilityVishal Rajpurohithttps://vishal.life/blog/llm-evaluation-harness-that-guides-releases/https://vishal.life/blog/llm-evaluation-harness-that-guides-releases/From task contracts and test slices to calibrated judges, regression budgets, and production feedback loops.Mon, 18 May 2026 00:00:00 GMTEvaluation and SecurityevalsLLM testingqualityrelease engineeringVishal Rajpurohithttps://vishal.life/blog/hallucination-controls-built-into-the-system/https://vishal.life/blog/hallucination-controls-built-into-the-system/A layered design for constraining claims, grounding evidence, verifying outputs, and abstaining when the system does not know.Sat, 09 May 2026 00:00:00 GMTEvaluation and SecurityhallucinationsgroundingverificationreliabilityVishal Rajpurohithttps://vishal.life/blog/guardrails-as-policy-enforcement/https://vishal.life/blog/guardrails-as-policy-enforcement/Building layered input, action, and output controls that remain testable when language and threats change.Thu, 30 Apr 2026 00:00:00 GMTEvaluation and SecurityguardrailspolicysafetymoderationVishal Rajpurohithttps://vishal.life/blog/threat-modeling-ai-applications/https://vishal.life/blog/threat-modeling-ai-applications/Assets, trust boundaries, model-specific attacks, tool abuse, data leakage, and concrete mitigations for deployed AI systems.Tue, 21 Apr 2026 00:00:00 GMTEvaluation and SecurityAI securitythreat modelingleast privilegedata protectionVishal Rajpurohithttps://vishal.life/blog/prompt-injection-defense-in-depth/https://vishal.life/blog/prompt-injection-defense-in-depth/Why delimiters are not a sandbox, how indirect injection reaches agents, and which architectural controls actually reduce impact.Sun, 12 Apr 2026 00:00:00 GMTEvaluation and Securityprompt injectionagentssecuritytool safetyVishal Rajpurohithttps://vishal.life/blog/embedding-systems-models-indexes-drift/https://vishal.life/blog/embedding-systems-models-indexes-drift/Selecting representations, constructing embedding inputs, migrating indexes, and detecting when semantic neighborhoods stop serving the product.Fri, 03 Apr 2026 00:00:00 GMTRetrievalembeddingsvector searchsemantic searchdriftVishal Rajpurohithttps://vishal.life/blog/fine-tuning-lora-and-adapter-operations/https://vishal.life/blog/fine-tuning-lora-and-adapter-operations/When fine-tuning is justified, how low-rank adapters work, and what it takes to evaluate, serve, and update them safely.Wed, 25 Mar 2026 00:00:00 GMTInference Systemsfine-tuningLoRAadapterstraining dataVishal Rajpurohithttps://vishal.life/blog/multimodal-ai-pipelines-images-audio-documents/https://vishal.life/blog/multimodal-ai-pipelines-images-audio-documents/Engineering preprocessing, temporal and spatial grounding, context budgets, validation, and storage around multimodal models.Mon, 16 Mar 2026 00:00:00 GMTMultimodal and Edgemultimodalvisionaudiodocument AIVishal Rajpurohithttps://vishal.life/blog/llm-cost-engineering-unit-economics/https://vishal.life/blog/llm-cost-engineering-unit-economics/Modeling full request cost, eliminating waste, forecasting margins, and optimizing without disguising quality regressions.Sat, 07 Mar 2026 00:00:00 GMTInference SystemsLLM costunit economicsoptimizationFinOpsVishal Rajpurohithttps://vishal.life/blog/observability-for-llm-and-agent-systems/https://vishal.life/blog/observability-for-llm-and-agent-systems/Tracing model calls, retrieval, tools, state transitions, quality signals, and cost without turning telemetry into a privacy liability.Thu, 26 Feb 2026 00:00:00 GMTProduct ArchitectureobservabilitytracingagentsoperationsVishal Rajpurohithttps://vishal.life/blog/model-routing-cascades-and-fallbacks/https://vishal.life/blog/model-routing-cascades-and-fallbacks/How to choose models per request using task risk, calibrated confidence, operational health, and total expected cost.Tue, 17 Feb 2026 00:00:00 GMTProduct Architecturemodel routingcascadesfallbacksinferenceVishal Rajpurohithttps://vishal.life/blog/structured-outputs-beyond-valid-json/https://vishal.life/blog/structured-outputs-beyond-valid-json/Designing schemas, constrained decoding, semantic validators, repairs, and safe evolution for dependable model integrations.Sun, 08 Feb 2026 00:00:00 GMTProduct Architecturestructured outputsJSON SchemavalidationAPI designVishal Rajpurohithttps://vishal.life/blog/model-context-protocol-production-architecture/https://vishal.life/blog/model-context-protocol-production-architecture/Designing MCP servers, capability discovery, authorization, transport, schemas, and observability for real agent ecosystems.Fri, 30 Jan 2026 00:00:00 GMTAgentsMCPtoolsprotocolsagentsVishal Rajpurohithttps://vishal.life/blog/edge-llm-inference-on-device/https://vishal.life/blog/edge-llm-inference-on-device/Quantization, memory, thermal budgets, runtimes, privacy, and hybrid execution for useful models on constrained hardware.Wed, 21 Jan 2026 00:00:00 GMTMultimodal and Edgeedge inferencequantizationon-device AIprivacyVishal Rajpurohithttps://vishal.life/blog/ai-data-flywheels-without-feedback-pollution/https://vishal.life/blog/ai-data-flywheels-without-feedback-pollution/Turning product use into better data through instrumentation, review, provenance, counterfactuals, and controlled training loops.Mon, 12 Jan 2026 00:00:00 GMTRetrievaldata flywheelfeedbacktraining dataML operationsVishal Rajpurohithttps://vishal.life/blog/ai-product-ux-for-uncertain-systems/https://vishal.life/blog/ai-product-ux-for-uncertain-systems/Designing intent, progress, evidence, editing, confirmation, and recovery so capability becomes trustworthy product behavior.Sun, 28 Dec 2025 00:00:00 GMTProduct ArchitectureAI UXproduct designtrustagentsVishal Rajpurohithttps://vishal.life/blog/laravel-php-production-ai-architecture/https://vishal.life/blog/laravel-php-production-ai-architecture/Queues, streaming, typed provider boundaries, retrieval, tool execution, and observability for durable AI features in Laravel.Tue, 16 Dec 2025 00:00:00 GMTProduct ArchitectureLaravelPHPAI engineeringqueuesVishal Rajpurohit