Fine-tuning & EvalsGuardrailsSecurityHourly
Red-Team & Harden a Customer-Facing Agent
Aegis Research · Remote (Global)
$120 / hr
The outcome
Before their agent goes public, they need someone to break it — prompt injection, jailbreaks, data leaks — and then build the guardrails.
The client
Aegis Research — Remote (Global). A fine-tuning & evals build we delivered end to end.
The challenge
Adversarially test a production LLM agent, document repeatable failures, and implement input/output guardrails and an eval suite that keeps them fixed.
What we built
- Probe for prompt injection, jailbreaks, and data exfiltration
- Implement input/output guardrails and safe-completion policies
- Add a red-team eval suite that runs on every prompt change
The stack
Stack: LLM security know-how, a guardrails library, and eval tooling.
Want something like this?
Tell us what you're building and we'll scope it — most projects start within a week.