Pierre KasparianAI & Data freelancer

Blog

Tech watch, experience reports and tutorials on generative AI.

AI agentsLLM pipelineAI agent orchestrationAI document assistant enterpriseLLM
Perplexity Brain: AI Agents That Learn From Their Own Work

Perplexity's Brain builds a self-improving context graph for AI agents: nightly synthesis, 25% accuracy boost on known tasks, 13% lower cost.

June 21, 2026 · 5 min read

RAGGDPRAI agentsGDPR-compliant AI integrationGDPR-compliant LLM deployment
MosaicLeaks: How RAG Agents Leak Your Private Data

Research agents combining private documents with web search leak confidential data through their queries. The PA-DR method cuts leakage from 34% to 9.9%.

June 21, 2026 · 7 min read

GDPRconsentGDPR-compliant AIcomplianceAI GDPR SMB Europe
Elkjop: €1.8M GDPR Fine for Forced Consent

Elkjop fined €1.8M by Norway's DPA for tying club membership to marketing consent. What this ruling means for European companies deploying AI systems.

June 19, 2026 · 6 min read

MCPAI AgentsAndroidLLM orchestrationenterprise AI tools
Android 17 Adopts MCP: What It Means for Your AI Agents

Android 17 implements MCP natively via AppFunctions. AI agents can now orchestrate any mobile app directly on-device, without routing data through cloud servers.

June 17, 2026 · 4 min read

AI AgentsAI integration costsAnthropicClaudeLLM orchestration
AI Agent Costs: What the Claude SDK Billing Crisis Revealed

Anthropic nearly billed its Agent SDK at API rates in June 2026. The episode reveals the real costs of production AI agents and how to budget for them as an SME.

June 17, 2026 · 5 min read

ClaudeAI engineeringLLM integrationAI agentsdeveloper productivity
80% of Anthropic's Code Written by Claude: Key Takeaways

In May 2026, 80% of Anthropic's production code was generated by Claude. Here are the 3 concrete steps to help your team reach a comparable level.

June 12, 2026 · 6 min read

RAGfine-tuningLLMGDPR-compliant AILLM pipeline Python
RAG vs Fine-tuning: Key Differences and When to Choose

RAG or fine-tuning to adapt an LLM to your domain? Key differences, use cases, costs, and GDPR implications for European companies.

June 5, 2026 · 7 min read

AI agentLLM pipeline Python freelanceorchestration LLM agentsRAG chatbot custom documents
AI agents vs LLM pipelines: when to choose which

Which architecture for your AI project: autonomous agent or fixed LLM pipeline? Practical comparison with examples to decide based on your use case.

June 5, 2026 · 6 min read

AI agentsovereign AI EuropeGDPR-compliant AI integrationenterprise AI governanceAI GDPR SMB Europe
Enterprise AI agents: governance is the real bottleneck

85% of enterprises run AI agent pilots, but only 5% reach production. The problem is not the model: it is permissions, identity, and GDPR compliance.

June 5, 2026 · 8 min read

MCPAI AgentsSecurityGDPROpenAI
Secure MCP Tunnel: Connect AI Agents to Private Systems

OpenAI's Secure MCP Tunnel lets AI agents access internal tools without exposing your servers to the internet. Architecture breakdown and GDPR analysis.

May 28, 2026 · 6 min read

RAGagentic RAGGDPR-compliant RAG productionLLM pipelineenterprise AI
RAG Is Not Dead: The Rise of Agentic RAG in Enterprise

Long-context models haven't killed retrieval-augmented generation. Here's why RAG remains essential for enterprises in 2026 and where it's heading.

May 28, 2026 · 6 min read

AI agentsRAG productionvector databaseElasticsearchLLM pipeline
Persistent AI Agent Memory with Elasticsearch

Architecture for multi-index AI agent memory on Elasticsearch: 3 memory types, hybrid retrieval, DLS multi-tenant isolation, and R@10 of 0.89. Production guide.

June 19, 2026 · 7 min read

LLMGDPR-compliant LLM deploymentLLM hosting EuropeAI cost optimizationLLM orchestration Python
LLM Inference Engineering: Optimize Latency and Costs

How LLM inference engineering works in production: prefill, decode, batching, quantization, and when to self-host to control costs and stay GDPR-compliant.

June 16, 2026 · 8 min read

AI agentLLM evaluationLLM orchestration Pythonfine-tuningRAG production
Evaluating an AI Agent in Production: the Semantic Judge

How to evaluate AI agent quality in production with a fine-tuned semantic judge: LangChain + Fireworks method, 100x cheaper than frontier LLMs.

June 16, 2026 · 7 min read

AI agentsOpikobservabilityLLM orchestrationproduction
AI Agent Harness: How to Make It Self-Repairing

Opik connects traces, automatic diagnosis and test loops: every production incident becomes a permanent regression test. Python examples.

June 12, 2026 · 7 min read

local LLMGDPRopen source LLM GDPR compliantLLM hosting EuropeLLM pipeline Python freelance
Local LLMs and GDPR: agentic coding without data leaks

Local LLMs or Mistral cloud for GDPR-compliant agentic coding: comparing sovereign architectures, practical guide with Ollama, LM Studio and VS Code.

June 5, 2026 · 7 min read

RAGRAG chatbot custom documentsGDPR-compliant RAG productionLLM pipeline Python freelanceretrieval augmented generation
Indexing images in a RAG pipeline: production guide

Images in a RAG pipeline: index-time captioning, junk image filtering, separate chunks. Results: 1-6% overhead instead of 27-51%.

June 5, 2026 · 7 min read

RAGevaluationproductionGDPR-compliant RAG productionLLM
7 Advanced Metrics to Evaluate Your RAG in Production

Standard metrics miss up to 40% of RAG errors in production. Discover 7 advanced evaluation techniques to detect hidden accuracy gaps in your pipeline.

May 28, 2026 · 7 min read

RAGchunkingNLPLLM pipeline Python freelanceLangChain
RAG Chunking: 4 Strategies to Maximize Retrieval Precision

Fixed-size, recursive, semantic or agentic: comparing 4 RAG chunking strategies with code examples and production recommendations.

May 28, 2026 · 8 min read

PythonPDFLiteParseRAGGDPR
LiteParse v2.0: Local PDF Extraction Without LLM or Cloud

LiteParse v2.0 parses PDFs and Office documents locally, without LLMs or cloud APIs. GDPR-compliant RAG pipelines in Python, JS, or Rust.

May 28, 2026 · 7 min read

LLMMistralLLM pipeline Python freelanceLLM cost optimization for SMBssovereign AI Europe freelance
Dynamic LLM Routing: Cheaper, Reduce Downtime

Routing across Mistral Small/Medium/Large based on token volume and server load can cut LLM costs by 10x with no quality loss. Here is the playbook.

May 28, 2026 · 9 min read

RAGrerankercross-encoderGDPR-compliant RAG productionLLM pipeline Python freelance
Boosting a RAG with a Cross-Encoder Reranker

A cross-encoder reranker improves RAG precision without changing your retriever. Cohere Rerank, local hosting options, Python examples.

May 28, 2026 · 7 min read

PythonPDFPyMuPDFRAGNLP
Parsing PDF documents with PyMuPDF in Python

A complete PyMuPDF (fitz) tutorial: text extraction, metadata, images, and structured blocks from PDFs. Perfect for building a RAG pipeline.

May 27, 2026 · 10 min read

LLMGDPRComplianceSovereign AIEU hosting
Integrating an LLM without violating GDPR: 2025 guide

Complete guide for EU companies: which GDPR articles apply to LLMs, why the Cloud Act is a problem, and which architectures keep you compliant.

January 15, 2025 · 8 min read