AI Engineer — Agent & Runtime Engineering

SecNinjaz Technologies LLP
Openings: 1 | Experience: around 2 years
About the role
Build the software that enables AI agents to carry out cybersecurity workflows reliably: orchestration, tool
integration, context, state, execution controls and model deployment.
You will develop and test AI systems in Python, working with cybersecurity specialists and a data/model
engineer. You need strong applied AI engineering and practical cybersecurity ability to understand
security tasks, interpret evidence and diagnose unsuccessful runs.
What you will do
- Build LLM applications and agent workflows using APIs, retrieval, structured outputs and tool calling.
- Develop the agent runtime: task orchestration, persistent state and context, retries, recovery,
cancellation and execution traces.
- Integrate security tools and enforce permissions, approval steps, scope boundaries, execution limits and
stop controls in application code. Handle credentials, untrusted target/tool content and isolated
execution safely.
- Integrate hosted and open-weight models while preserving application-owned state, evidence and tool
contracts. Adapt prompts and integrations where needed and test for regressions.
- Capture task context, tool actions/results, failures, versions, reviewer corrections, timing and cost in a
form the data/model engineer can use for reviewed datasets and evaluations.
- Jointly define versioned trace and tool interfaces, evaluation criteria and release checks with the
data/model engineer and cybersecurity specialists. Security specialists review domain labels and
validate findings.
- Support observable, recoverable workloads and deployment, including private or self-hosted
environments; investigate concurrency, reliability and resource-use problems with the engineering
team.
- Work with security specialists and the data/model engineer to test task success, evidence quality, false
positives, latency and cost, and release improvements with regression checks and rollback.
SecNinjaz Technologies LLP 1What you should bring
- Around two years of hands-on development experience; strong Python, APIs, Git, debugging and
automated testing.
- A working AI/LLM application or agent with tool integration, state, meaningful tests and failure handling.
Be ready to explain, modify and debug your own contribution.
- Practical understanding of model limitations, prompting, structured outputs, retrieval and evaluation.
- Conceptual understanding of training versus inference and of reinforcement-learning fundamentals:
environments, actions, rewards and evaluation. An implemented RL project is welcome but not required
for this role.
- Demonstrated practical cybersecurity ability: reason about authentication, authorization, trust
boundaries and common web/API or code weaknesses; interpret tool output, reproduce and validate a
finding, reject a false positive and verify a fix in a controlled environment.
- Comfort with Linux, HTTP/APIs, logs and application deployment, and explaining technical decisions to
teammates.
Relevant professional work, labs, research and personal projects can demonstrate these skills.
Good to have
- Docker, CI/CD, queues, monitoring or production reliability experience.
- Private inference, open-weight model serving or inference performance work.
- Deeper security-tool integration, detection engineering or security research.
- Fine-tuning or implemented RL experiments.
Initial outcomes, with the team
- Deliver a tested security workflow with useful traces and independently validated outcomes.
- Demonstrate recovery from a failed run and a control, enforced in application code, that blocks an
unapproved action.
- Run the workflow through two model integrations with retained application state and regression checks;
supply usable trace records to the data/evaluation pipeline.
Selection process
A project discussion and a short practical exercise using a supplied lab or sanitized security case. You will
interpret tool outputs that include a false positive, diagnose a failed agent run, implement a recovery or
permission-control improvement, and define a trustworthy success check.
AI tools may be used; explain your contribution and be ready to debug your solution. Nothing is run
against real systems.
All security work at SecNinjaz is authorized and operates under agreed rules of engagement.

Work arrangement
No