Door categorieën bladeren
Ontdekken
Fiverr Pro
Nederlands
$
USD
I will build llm evaluation tests and red team ai app for prompt injection
Muhammad A
Over deze dienst
Shipping an AI feature without tests usually goes one of two ways: it quietly gets worse after a prompt change, or a user talks it into something it shouldn't do. I build the tests that catch both.
I'm a developer with 3 years of experience building AI apps. I test chatbots, RAG systems and agents built on OpenAI, Claude, Gemini or open models.
What you get
- A test set built from your real use cases, with the answers you expect
- Automated grading for accuracy, groundedness, tone and format
- Red-team tests for prompt injection, jailbreaks, system prompt leaks and data leaks, mapped to the OWASP Top 10 for LLM Applications
- A report ranked by severity, with steps to reproduce and suggested fixes
- On Premium, tests that run in your CI with promptfoo, Langfuse or Braintrust, so every prompt or model change gets scored
I test on staging or a copy of your app, never against real users. You pay your own API costs, which I estimate first.
No test suite can promise an app that never makes mistakes. What it gives you is a number you can watch and a list of what to fix first.
Send me what your AI feature does, a way to try it, and the failures you worry about most.
Maak kennis met Muhammad A
Muhammad A
arehman
- Afkomstig uitPakistan
- Lid sindsjan 2024
- Laatste levering5 maanden
Talen
Engels
I'm a software developer with 3+ years of experience, mostly designing and building large scalable software systems, AI apps and trading bots. That work taught me where things break once real users or real money are involved, so that's what I test for. Here I set up AI agents and local LLMs, build n8n, Make and Zapier automations, make chatbots, autonomous systems and voice agents that answer from your own content, fix apps built with Lovable, Bolt or Base44, and code trading bots in Pine Script, MQL5 and Python. You get the source code, documentation. Reach out and tell me what you need.
Andere AI-development diensten die ik aanbied
Veelgestelde vragen
Do you need my source code?
Not for Basic; a test URL or API access is enough. Standard and Premium need repo access to add the tests to your project.
Which tools do you use?
promptfoo, Langfuse, Braintrust or Ragas, depending on your stack. If you already use one, I build on it.
Will this stop all hallucinations?
No test can. It measures how often they happen, shows where, and checks that each fix actually helps.
Who pays for the API calls the tests make?
You do, on your own keys. I estimate the cost before we start and keep test runs small.
Can you test agents that use tools?
Yes. I check that the agent picks the right tool, passes the right inputs and refuses unsafe actions. Tool-heavy agents fit Premium or the agent extra.
Can you fix the problems you find?
Premium includes fixes to prompts, retrieval settings and guardrails, plus a retest. On Basic and Standard, fixes can be added through a custom offer.
Will you keep my data private?
Yes. I use test data where possible, sign an NDA if you need one and delete copies of your data when we're done.
