AI Engineering

Kaan Un

Evaluation-first AI systems with model telemetry, guardrails, and trace-driven iteration. Built by Kaan Un, founder of Traction IT.

Shader Layer Active

WebGL background and atmospheric motion load only on this route group.

System Performance

Benchmark Dashboard

Model Accuracy

Eval Coverage

Latency Reduction (ms)

System Architecture

Pipeline Overview

DataEmbedRetrieveGenerateEvaluateDeploy

Eval Pulse

System Metrics

0+

Evals Implemented

Offline and online evaluations deployed.

0%

Latency Improvement

Reduction in end-to-end inference latency.

0+

Model Iterations

Controlled experiments completed with trace instrumentation.

0%

Guardrail Coverage

Core user flows covered by safety and quality checks.

Model + Data + Delivery

Applied Skills

ML and Evaluation

  • LLM Evaluation Design

    Creates offline and online evals for quality and safety.

  • Error Analysis

    Uses traces to identify failure clusters and root causes.

  • RAG Systems

    Builds retrieval pipelines with observability and guardrails.

Platform Engineering

  • Model Serving

    Deploys low-latency endpoints with fallback strategies.

  • Telemetry

    Instruments LLM workflows for debugging and monitoring.

  • Synthetic Data

    Generates focused test sets for edge cases and regressions.

Latest Publications

Technical Notes

The Real AI Gold Rush Is Accelerated Data Enrichment

2026-03-14

The Real AI Gold Rush Is Accelerated Data Enrichment

The real alpha is not in gradient apps or vibe-coded clones. It is in accelerated data enrichment and AI-native quantitative research.

20 likes3 comments
aidata-enrichmentquant-finance
Read more
One Tool, One Command, Zero Friction Between Insight and Delivery

2026-03-09

One Tool, One Command, Zero Friction Between Insight and Delivery

I burned through my remaining credits to deploy an agent skill that generates executive reports with live charts and KPIs.

40 likes5 comments
agent-skillsautomationreports
Read more
Platform Lock-In Destroys Agentic Coding Efficiency

2026-03-06

Platform Lock-In Destroys Agentic Coding Efficiency

Agent skills are token optimized and highly efficient, but they break when you move to a new ecosystem. Platform-agnostic junctions fix that.

15 likes1 comments
agent-skillsplatform-agnosticcursor
Read more
Stress Testing Reddit, Bluesky, and X for Quant Trading Signals

2026-03-04

Stress Testing Reddit, Bluesky, and X for Quant Trading Signals

Three platforms, three distinct approaches, one clear winner that captures actual market signal instead of speculative noise.

25 likes3 comments
quant-financesocial-signalsdata-analysis
Read more
A 2-Hour Cursor Deep Dive Changed How I Think About Education

2026-02-28

A 2-Hour Cursor Deep Dive Changed How I Think About Education

For the 9th instalment of Cursor Meetup in Canada's Capital, I transformed the format from surface level AI exploration into automation mechanics.

63 likes15 comments
cursoreducationworkshops
Read more
The Interface Layer for Autonomous Systems Is Collapsing Into Tool Calls

2026-02-25

The Interface Layer for Autonomous Systems Is Collapsing Into Tool Calls

Terminal native AI agents now execute decisions, manipulate infrastructure, and manage entire systems without human intermediation.

25 likes
autonomous-systemsterminal-agentstool-calls
Read more
Read all 38 posts →

Recent Events

Lab Sessions

View Full Resumes