About

A research team that builds first, then writes

KP Journal started in 2023 as an internal knowledge base for a Cambridge-based AI consulting group. We started publishing because we kept finding the same gaps in public writing about AI agents — either too high-level or too product-specific. This is the middle ground we wish had existed.

Team working at a table with laptops
What we do and why

How this publication operates

KP Journal is an independent research publication. The advisory practice — process assessments, technical reviews, and technical publications for engineering audiences — funds the research and keeps it free of sponsorship or advertising dependencies.

Editorial content is not written to support service sales, and we do not accept paid placements or vendor-directed coverage. When we assess a tool or architecture, it is because we have used it in an engagement or tested it directly. Where links to third-party tools carry a referral arrangement, this is noted at the content level.

Get in touch
How we work

A few principles we try to hold to

No synthetic expertise

We write about things we've actually tested or implemented. If something is speculative, we say so. We don't write "how AI agents work" articles sourced from other people's "how AI agents work" articles.

Editorial independence

Tool references in our articles reflect direct use, not placement. Where a referral arrangement exists, it is noted inline. Advisory engagements do not direct editorial coverage — clients are not identified in articles without explicit permission.

Reproducible findings

When we make a claim about a model's behavior or an architecture choice, we describe the conditions and the test. You should be able to replicate or contest it.

The team

Who writes here

Our team combines backgrounds in software engineering, ML research, and technical writing.

James Whitfield

James Whitfield

Editor & AI Systems Researcher

Editor and AI systems researcher. Background in distributed systems; has spent four years designing multi-agent pipelines for document processing and customer operations. Writes on agent architecture, tool-calling reliability, and production failure modes.

Clara Novak

Clara Novak

Automation Strategist

Operations consultant specializing in workflow analysis and automation feasibility. Leads process readiness assessments. Has worked across logistics, insurance, and professional services. Writes on the operational conditions that determine whether automation succeeds.

Daniel Reeves

Daniel Reeves

Prompt Engineering Lead

Former NLP researcher focused on instruction-tuned models. Designs prompt evaluation frameworks; writes on reliability, structured output generation, and the gap between benchmark performance and production behavior.

Sophie Laurent

Sophie Laurent

Technical Content Director

Technical Content Director. Has written for engineering audiences in cloud infrastructure, security, and applied ML. Oversees the white paper and long-form publication practice.

Tomasz Kowalski

Tomasz Kowalski

Infrastructure & Evaluation Engineer

Infrastructure and evaluation engineer. Builds internal tooling for agent benchmarking and prompt reliability measurement. Contributes technical appendices and reproducibility documentation.

Amara Osei

Amara Osei

Research Analyst

Research analyst. Tracks developments in agent memory architectures, multi-model orchestration, and retrieval-augmented systems; produces structured analysis for longer research pieces.

From the work

A few glimpses

Research sessions, whiteboard diagrams, and the kind of desks where this stuff actually gets written.

Team discussion at a whiteboard
Laptop with code on a wooden desk
People collaborating at a table
Notebook with technical notes
Data visualization on a monitor
Person working on a laptop in a bright office