Sfoglia categorie
Esplora
Fiverr Pro
Italiano
$
USD
I will build an llm evaluation harness with guardrails and a ci quality gate
Jigon Y
Informazioni su questo servizio
Your LLM app works in the demo. The question is what ships to users after the next prompt change - and most teams have no way to know. That is what I build: the LLM evaluation layer for your feature.
What you get:
- Eval harness: a golden set of labeled test cases for your pipeline, scored automatically. Exact checks where exact is fair, judge scoring where wording varies - so a correct answer phrased differently still passes.
- CI quality gate (Standard+): the harness runs on every change and fails the build when quality regresses. Silent regressions stop shipping.
- Guardrails (Advanced): an input guard against prompt injection and jailbreaks, an output guard that redacts leaked secrets and PII - each measured with its own committed test battery, including false-positive checks on benign lookalikes. A guard that blocks real users is just a different kind of outage.
How I work: honest numbers only. Every metric I report is reproducible from committed test cases - you can rerun everything yourself. If your setup will not benefit from this, I say so before you pay, not after.
Works with OpenAI, Claude, Gemini or your own models. Python-based, integrates with GitHub Actions or any CI.
Scopri di più su Jigon Y
Jigon Y
Data and Automation Engineer, Python, Web Tools, Clean Data
- DaCorea del Sud
- Membro dalug 2026
- Tempo di risposta medio2 ore
Lingue
Coreano, Inglese
I'm a data engineer who builds practical tools and turns messy data into clean, usable results. Fast, clear, reliable.
What I do:
- Python automation and scripts (scraping, cleaning, Excel/CSV, reports)
- Custom web tools, calculators and widgets (React/HTML)
- Data cleaning, formatting and light analysis
- Korean/English translation
I use AI to move faster, then personally review, test, and stand behind every deliverable. You get speed plus a human who owns the result.
Tell me your goal and I'll say honestly if I'm a fit.
Il mio portfolio
Altri servizi della categoria Sviluppo AI offerti da me
FAQ
Do I need to share my API keys or codebase?
For Starter, no - I can build the harness against sample inputs and outputs you provide. CI integration needs repo access or a sandbox. Keys stay yours: use a limited test key, or your team runs the live calls.
Which models and frameworks do you support?
OpenAI, Claude, Gemini or your own models. The harness is plain Python + Pytest, so it works with any stack and any CI - GitHub Actions, GitLab CI, Jenkins. No framework lock-in.
What numbers will I actually get?
Pass/fail per test case, aggregate scores per run, and for guardrails: block rate plus false-positive rate on a committed battery. Every number is reproducible - rerun the suite yourself and get the same result.
