AI Prompt Engineer Skills And Prompts
Pixalate
A complete application in a minute
Tailored resume and cover letter, ready to send.
Benefits offered by this job
Job summary
Pixalate is hiring an AI Prompt Engineer to own the production agent layer, writing and maintaining system prompts and agent skills that guide IVT and analytics agents. You’ll measure and version changes, collaborate with the IQ Agent and MCP teams, and ensure safe, accurate agent behavior in production.
The role emphasizes practical engineering, tool design, cost-aware tuning, and strong English writing, with a remote setup in Pakistan and U.S. timezone overlap for meetings.
Qualifications
- Bachelor's in Computer Science, AI, ML, or related field (or exceptional research track)
- Demonstrable experience building LLM agents that shipped to real users
- Hands‑on work with agent skills, plugins, system prompts, or equivalent agent‑behaviour tooling
- Evidence of evaluation discipline: you can describe an eval suite you built, the ground truth behind it, and a regression it caught
- Strong written English. This role is writing, and the writing is the product
- Working proficiency in Python or TypeScript, and comfort reading the codebase your agents run in
- Understanding of tool‑calling / function‑calling and how agents select tools
- Awareness of prompt injection, PII handling, and the failure modes of customer‑facing AI
- Comfortable with some U.S. timezone overlap for team meetings
- Ad‑tech, digital advertising, or fraud‑detection exposure is strongly preferred
- Experience with MCP (Model Context Protocol) servers or similar tool‑integration standards is a plus
- Experience with observability and tracing for LLM systems is a plus
- Experience with retrieval systems and their failure modes is a plus
- Data analysis background, including SQL and working with large datasets, is a plus
Responsibilities
- Build and maintain the system prompts and agent skills that define agent behavior
- Own the design of when agents ask clarifying questions rather than guessing
- Version prompts and skills with clear change records
- Debug real agent failures and reduce hallucinations and context exhaustion
- Design context strategies and tune tool surfaces for cost, latency, and accuracy
Skills
Education
Tools
Job description
Job Description:
Employment Type: Full-Time
Location: Remote, Pakistan
Pixalate runs AI agents in production against real customer problems: agents that investigate invalid traffic (IVT) for clients, and analytics agents that let customers ask questions of our data directly. These agents are only as good as the instructions and skills behind them.
As an AI Prompt Engineer you own that layer. You will write and maintain the system prompts and the agent skills that decide how our agents behave: what they know, which tools they reach for, when they push back, and, critically, when they stop and ask the user a clarifying question instead of guessing. This is engineering work, not prompt tinkering. You will version your changes, measure them against evaluations, and be accountable for agent behaviour in front of real customers. You will work directly with the team building the IQ Agent and Analytics MCP, the client IVT investigator agents, and the MRT MCP.
- Author and maintain Claude skills and plugins that define agent capabilities and behaviour
- Design agents that ask clarifying questions and gather requirements interactively, rather than answering one-shot from incomplete information
- Decide what belongs in a skill, what belongs in a system prompt, and what belongs in a tool
- Maintain the skill library as products evolve, treating it as a living surface rather than a one-time build
- Own the system prompts behind Pixalate's client-facing agents: IVT investigator agents, IQ Agent / Analytics MCP, and MRT MCP
- Version prompts and skills deliberately, with a clear record of what changed and why
- Debug real agent failures: wrong tool selection, hallucinated findings, missed clarifying questions, context exhaustion
- Define what "correct" means for each agent and build evaluation suites that test it
- Build ground-truth datasets from real customer scenarios and measure changes against them
- Catch regressions before customers do
- Engineer context strategies: what the agent sees, in what order, within what budget
- Work with engineering on MCP tool surfaces: which tools exist, how they are described, and how those descriptions drive correct selection
- Tune for cost and latency as well as accuracy
- Guard against prompt injection, data leakage, and unsafe tool use in customer-facing agents
- Ensure agents represent Pixalate's findings accurately and do not overstate confidence
- Bachelor's in Computer Science, AI, Machine Learning, or related field (or exceptional research track record)
- Demonstrable experience building LLM agents that shipped to real users, not prototypes or demos
- Hands‑on work with agent skills, plugins, system prompts, or equivalent agent‑behaviour tooling
- Evidence of evaluation discipline: you can describe an eval suite you built, the ground truth behind it, and a regression it caught. This is the single most important qualification.
- Strong written English. This role is writing, and the writing is the product.
- Working proficiency in Python or TypeScript, and comfort reading the codebase your agents run in
- Understanding of tool‑calling / function‑calling and how agents select tools
- Awareness of prompt injection, PII handling, and the failure modes of customer‑facing AI
- Comfortable with some U.S. timezone overlap for team meetings
- Ad‑tech, digital advertising, or fraud‑detection exposure is strongly preferred: programmatic, DSPs, SSPs, CTV, or IVT and ad fraud concepts. If you do not have it, you must be prepared to learn the domain deeply in your first 90 days, because agent quality depends on it.
- Experience with MCP (Model Context Protocol) servers or similar tool‑integration standards is a plus
- Experience with observability and tracing for LLM systems is a plus
- Experience with retrieval systems and their failure modes is a plus
- Data analysis background, including SQL and working with large datasets, is a plus
Pixalate is an online trust and safety platform that protects businesses, consumers and children from deceptive, fraudulent and non-compliant mobile, CTV apps and websites.
We're Seeking An AI Prompt Engineer To Help Build And Maintain The Production Agent Systems That Directly Impact Millions Of Users Worldwide. Our Software And Data Have Been Used To Unearth Multiple High Profile Criminal And Illegal Surveillance Cases Including
- UNICEF: Pixalate was recently recognized by UNICEF as a Key Innovator for protecting children’s online privacy. & UNICEF Global Compliance Tech case study
- Gizmodo: An iCloud Feature Is Enabling a $65 Million Scam, New Research Says
- Adweek: A 7-Figure Ad Fraud Scheme Running on Roku Underlines Murkiness of CTV
- Washington Post: Your kids’ apps are spying on them
- Pro Publica: Porn, Piracy, Fraud: What Lurks Inside Google’s Black Box Ad Empire
- ABC7 News: The State of Children's Privacy Online
- NBC News: How many apps are tracking your children
Our team of lawyers, data scientists, engineers, economists and researchers span globally with presence in California, New York, Washington DC, London and Singapore.
Pixalate is an equal opportunity employer committed to building a diverse team.
We believe in small teams that produce high output. Slack is a way of life and short emails are encouraged. We value a fearless attitude and bold ideas. Chess players do well here. Titles don't mean much - you attain respect by producing results. Everyone's a data addict and analytical thinker who thrives on collaboration.
- A Competitive Compensation Package
- Monthly Internet Reimbursement
- Casual, Remote Work Environment
- Flexible Hours
- Opportunity for Advancement and Professional Development
- Fun Annual Team Events
- Being part of a high-performing global team that wants to win and have fun doing it
'/%3e%3cpath%20d='M101.744%2012.4726L102.894%2016.1947L99.0021%2013.2201L101.744%2012.4726ZM105.3%208.40937H100.488L99.0026%203.59473L97.5116%208.41047L92.7%208.40385L96.5965%2011.3834L95.1055%2016.1942L99.0021%2013.2196L101.408%2011.3834L105.3%208.40992V8.40937Z'%20fill='white'/%3e%3cdefs%3e%3clinearGradient%20id='paint0_linear_5593_1381'%20x1='90'%20y1='9.44471'%20x2='99'%20y2='9.44471'%20gradientUnits='userSpaceOnUse'%3e%3cstop%20stop-color='%2300AD70'/%3e%3cstop%20offset='0.998594'%20stop-color='%2300B67A'/%3e%3cstop%20offset='1'%20stop-color='%23DCDCE6'/%3e%3c/linearGradient%3e%3c/defs%3e%3c/svg%3e){kind=small}
