AI engineering
AI engineering by D. Klincharski, focused on useful agent interfaces, measurable token efficiency, and answers grounded in source material. Explore MCP design notes, retrieval experiments, and the products that put these ideas to work.
About D. KlincharskiBuilds / Tools / Labs
Projects to explore.
01
Open source
ReelHiveTurn a brief into a narrated video with Strands graphs, a local editing studio, quality checks, and resumable runs.
02
Open source
AITOSet up token-efficient AI coding workflows and measure the results.
03
Product
RafiHiveSupport EU pharma research, dossier gap checks, and packaging reviews with cited evidence and qualified RA oversight.
Articles / Experiments
Notes from the work.
01
Oct 21, 2026
Amazon Bedrock RAG Architecture: 5 Ways to Build with Managed Knowledge BasesManaged vs customer-managed Knowledge Bases, then five Amazon Bedrock RAG architectures: direct inference, AgentCore Harness, Strands and Agentic Retrieval, with trade-offs.
02
Oct 11, 2026
Inside ReelHive: an agent harness built with Strands GraphHow six agent roles, Strands graphs, production scripts, and quality gates work together to turn a brief into a narrated video in ReelHive.
03
Oct 11, 2026
What building a video agent taught me about agent harnessesBuilding a video pipeline with Strands Graph changed where I put the effort: state, checks, recovery and approvals. Why I think agent harnesses come next.
04
Oct 05, 2026
RAG on AWS: Which Pattern Fits Your Use Case?Eight RAG patterns in one visual guide: practical examples, what AWS supports, and where retrieval is heading.
05
Oct 05, 2026
SDLC vs AI-DLC: do AI agents really need a new development lifecycle?Everyone is redrawing the software lifecycle for AI agents. A critical look at AWS AI-DLC and the "ADLC" infographics, what coding agents already do on their own, what holds up, what doesn't, and where this is heading.
06
Oct 05, 2026
What is Jev? Why everyone is excited about a model that can't write a sentenceJev from TypeSafe AI returns typed decisions instead of text. A critical overview of the hype, the vendor numbers, how it fits on AWS with Bedrock AgentCore, and where it is actually useful.
07
Sep 29, 2026
An SEO and GEO Runbook for Product Sites: What Actually Moved the NeedleA field runbook for getting a product site crawled, understood, and quoted by Google, ChatGPT, Perplexity, and Claude, with the pitfalls that cost the most time.
08
Sep 18, 2026
The AI boom has entered its compliance phase: what the EU AI Act changes for companiesFor years the AI question inside companies was what models could do. The EU AI Act turns it into something harder: can you prove what your AI systems access, what protects them and who is responsible?
09
Sep 17, 2026
Agentic AI in regulatory affairs: how much autonomy is too much?FDA is already deploying agentic AI, while pharma companies are exploring AI agents for regulatory intelligence and authoring. What should an AI agent actually be allowed to do in regulatory affairs?
10
Sep 02, 2026
Why AI gets regulatory affairs wrong when you remove the source documentsLanguage models are excellent at sounding like a regulation and unreliable at being one. Here is what goes wrong when the primary source is out of the loop, and how to keep it in.
11
Aug 09, 2026
Bedrock Knowledge Base Chunking Strategies: How to Choose (and What Changed in 2026)A practical decision guide to Bedrock Knowledge Base chunking, the managed-versus-customer-managed fork, starting parameters, evaluation, and Terraform.
12
Jul 28, 2026
MCP v2 Migration Guide: What Changed and Should You Upgrade?A practical MCP v2 migration guide covering SDK changes, the stateless 2026 protocol, compatibility risks, and when upgrading is worth it.
13
Jul 28, 2026
Should Every MCP Server Ship with an Agent Skill?An MCP server must stand on its own. Learn when an optional agent skill adds real value—and when it only hides a poor interface.
14
Jul 28, 2026
15 Rules for Designing Token-Efficient MCP APIs for AI AgentsFifteen practical rules for MCP APIs that reduce tool calls, context bloat, latency, and cost without starving AI agents of useful data.
15
Jul 24, 2026
Why domain expertise belongs in the product loopAI can accelerate regulated work, but useful software begins when domain specialists shape the product itself.
16
Jul 21, 2026
What RafiHive is for, and who actually needs itRafiHive supports EU pharma regulatory research, Module 1 gap checks, and packaging and labelling reviews with cited sources and qualified human oversight.
Explore by topic