01
Evaluation systems
Deterministic graders, realistic terminal tasks, pinned environments, and anti-cheat validation that preserve signal.
Software engineer · Salt Lake City
I’m Mel Kon. I build systems that can’t afford to guess.
Backend engineer focused on data pipelines, APIs, and cloud infrastructure, with an emphasis on observability, reproducibility, and operational reliability. Currently developing AI model evaluation systems at Snorkel AI
01 / Selected evidence
Work measured by what changed.
01
Snorkel AI — Frontier Model Evals
Built and reviewed software-engineering evaluations that measure whether frontier AI models can ship correct code.
50+ evaluations built200+ evaluations reviewed
02LAUSD Snowflake Identity Migration (Ping AIC)
Patched connector to enable Snowflake sync; migration unblocked.
250K+ identitiesBi-directional live sync (Snowflake ↔ Ping AIC)
03Georgia Courts — Docket Notices
Modernized docket notices to Kotlin with deterministic PDF generation.
Java → Kotlin rewriteDeterministic PDFs
Personal build · real-world outcome
An audio-first citizenship study app I built for my mom.
Bilingual, audio-first exam prep with adaptive testing—built for one real learner, and it helped her pass.
Read the story02 / Operating range
Useful where software meets messy reality.
The common thread is not a framework. It is turning ambiguous, failure-prone systems into something a team can test, observe, and trust.
02
Identity & data
Schema mapping, reconciliation, connector diagnostics, live sync, and careful production migration at enterprise scale.
03
Cloud tooling
Telemetry pipelines, backend APIs, observability, CI/CD, and reproducible systems that help teams move with confidence.
03 / Open channel
Have a difficult system that needs a calmer path forward?
I’m open to software engineering roles and focused contract work. A concise email is the fastest way to start.