Machine learning engineer

I build machine learning systems you can measure, trace, and improve.

I work across classification, retrieval, evaluation, data pipelines, and LLM products. I take ownership from problem definition through delivery, including the tests and traces needed to see what the model is doing.

See selected work Read my CV Email me

Selected work

Work I led from definition through delivery.

I define the problem, build the system, and show how it performs.

01

Sift

AI Engineer · Jun 2025 to Jun 2026

Making model behaviour visible from query to answer.

I owned how models classified questions, retrieved context, and exposed the evidence behind each answer. I also built the evaluation and access-control layers around that path.

What I owned

  • Led the observability layer for queries and answers, recording feedback, retrieved context, confidence, embedding activity, and latency.
  • Architected an organisation-specific semantic classifier with sentence-transformer embeddings, confidence thresholds, and deterministic fallbacks.
  • Linked AI answers to exact Slack and GitHub evidence through hybrid retrieval, decision traces, sufficiency gates, and private-channel controls.

02

Klysera

Research & Development Lead (Contract) · Sep 2025 to May 2026

Defining the research behind the Ideal Klysera Engineer.

I defined the intellectual property behind the Ideal Klysera Engineer, then built the research operation that connected the claim to curriculum, product positioning, and engineering career frameworks.

What I owned

  • Led four research programs, each with a defined question, evidence threshold, and review process.
  • Analysed 200 current engineering roles from 36 companies against a 50-role historical baseline using reproducible pandas pipelines.
  • Built a Next.js and Supabase analytics product for skill gaps, role coverage, taxonomy views, and curriculum decisions.

03

Cortex MIPs Working Group

Protocol Researcher · Jun 2026 to present

Testing MIP claims with reproducible experiments.

I turn claims in Monad Improvement Proposals into hypotheses, Testnet experiments, and automated test harnesses that other engineers can rerun.

View the MIP research lab

What I owned

  • Built a research lab that supports experiments across multiple MIPs.
  • Isolated an environment-dependent Foundry compatibility issue and turned it into a regression test.

04

Mono Connect.js

Independent documentation project · 2024

Turning Mono Connect.js into documentation developers can follow.

I turned a long package README into a searchable documentation site. It now covers the first working integration, framework setup, API references, reauthorisation, direct debit, and client/server boundaries.

Read the Mono Connect.js docs View the source

What I owned

  • Mapped the documentation to account linking, framework setup, events, methods, and advanced flows.
  • Built dedicated React, Angular, and Next.js integration guides instead of making developers translate one generic example.

Writing

More than 200 technical articles across software engineering, infrastructure, developer tools, and applied AI.

Current notes on AI systems and engineeringsammai.bearblog.dev Using Docker Init in Node.jsEarthly Why Fastify is a better Node.js framework than ExpressSoshace

Contact

Need an ML engineer who can own the model and the system around it?

samuelumoren365@gmail.com