AI Agent Engineer

Manifest Global; Delhi; On-site; INR 450,000 - 700,000; Full time

Job summary

Manifest Global is seeking a senior engineer to architect and ship production AI agents across its brands. You will own the end-to-end design, including model selection, tools, memory, guardian rails, and orchestration, turning individual agents into a scalable platform.

You will define success metrics, optimize latency and cost, and build safety and evaluation infrastructure. The role requires translating complex user problems into concrete agent specifications, with impact across counselling,

Responsibilities

• Design and build production AI agents across Manifest's brands, starting with high-leverage workflows in counselling, admissions outreach, and operational triage.
• Own the architecture decisions: model selection, tool use, memory, evaluation, guardrails, and the orchestration layer that holds it together.
• Build the abstractions that turn one good agent into a platform.
• Define what "working" means for each agent in measurable terms and build the evaluation infrastructure to prove it.
• Own latency, cost, and accuracy as engineering disciplines, not afterthoughts.
• Build the safety layer: when an agent should stop, elevate, or ask.
• Translate fuzzy user problems into agent specifications that can actually be built.
• Close the loop between user feedback and model behavior faster than anyone in this industry currently does.
• Set the standard for how AI agents are built at Manifest tooling, frameworks, model providers, evaluation harness, and deployment patterns.
• Stay close to the research frontier to know what is becoming possible six months out, and ship accordingly.
• In the first weeks, you’ll define high-leverage workflows and have something in the hands of a real user by month three.
• By mid-period, at least one agent is in production and measurably better than the manual workflow it replaced.
• Longer term, the agent layer is infrastructure, not a project.

Job description

The Agent Platform

The candidate will have responsibilities across the following functions

• Design and build production AI agents across Manifest's brands, starting with the highest-leverage workflows in counselling, admissions outreach, and operational triage.
• Own the architecture decisions: model selection, tool use, memory, evaluation, guardrails, and the orchestration layer that holds it together.
• Build the abstractions that turn one good agent into a platform.

The Agent Platform

The candidate will have responsibilities across the following functions

• Design and build production AI agents across Manifest's brands, starting with the highest-leverage workflows in counselling, admissions outreach, and operational triage.
• Own the architecture decisions: model selection, tool use, memory, evaluation, guardrails, and the orchestration layer that holds it together.
• Build the abstractions that turn one good agent into a platform.

Evaluation And Reliability

• Define what "working" means for each agent in measurable terms, and build the evaluation infrastructure that proves it.
• Own latency, cost, and accuracy as engineering disciplines, not afterthoughts.
• Build the safety layer: when an agent should stop, when it should elevate, when it should ask.

Integration Into The Product

• Work directly with product and engineering teams across Cialfo, BridgeU, Kaaiser, and Explore to ship agents into the surfaces where users already are.
• Translate fuzzy user problems into agent specifications that can actually be built.
• Close the loop between user feedback and model behaviour faster than anyone in this industry currently does.

Technical Direction

• Set the standard for how AI agents are built at Manifest tooling, frameworks, model providers, evaluation harness, and deployment patterns.
• Stay close enough to the research frontier to know what is becoming possible six months out, and ship accordingly.

What Success Looks Like

• The markers below reflect where Manifest's AI agent function is today. We'll calibrate the specifics once you're in the seat. These are directional, not fixed.
• In the first weeks, you have a clear point of view on which workflows across the portfolio are the highest-leverage places to deploy agents, and you can defend that view to the counsellors, product leaders, and engineers who will use what you build. You have started building, not just planning. Something is in the hands of a real user by the end of month three.
• By mid-period, at least one agent you have built is in production, used daily, and measurably better than the manual workflow it replaced. The evaluation infrastructure exists and is trusted by the people whose work depends on it. Other engineers across the group know how to build agents the Manifest way because you have made that pattern legible.
• Longer term, the agent layer is infrastructure, not a project. New product surfaces ship with AI agents as a default, not an exception. The work you built outlasts the specific models you started with, because the architecture you chose absorbed model upgrades without rewriting from scratch. The specifics will be calibrated once you're in the role. The direction won't change.

Requirements

• You have spent the better part of your career building software that runs in production, and somewhere in the last two years, you became one of the people other engineers go to when they have a hard question about LLMs, agents, or how to actually make this stuff work outside a notebook. You have shipped something agentic. Not a prototype. Something users use.
• You know the current frontier well enough to be opinionated about it. You have working views on which model providers to use for what, which orchestration patterns hold up under real load, where the abstractions in popular agent frameworks leak, and how to evaluate something that doesn't have a single correct answer. You write production-quality Python. You are comfortable with the surrounding stack: vector stores, prompt management, evaluation frameworks, and observability tooling.
• You think in systems, not in prompts. When you see a workflow, you instinctively map where reasoning happens, where state lives, where tools get called, and where humans need to stay in the loop. You build for the model that ships in six months, not the one that exists today.
• You are self-directed in the way this kind of role requires. There is no finished playbook for what you are building. The questions you will face, which agent to build next, how much to invest in evaluation infrastructure, when to use a smaller model and when to use a frontier one, will not have obvious answers. You make calls, you ship, you learn, you adjust.
• Somewhere underneath the technical depth, you care about what the agents you build actually do in the world. The fact that a student in Delhi, Manila or Lagos is going to make a better university decision because of the code you wrote matters to you.
• Most importantly, you read the description of what Manifest is building in AI, and your first reaction was that the agents I want to build are exactly the ones this company needs built. That's the person this role is for.

This job was posted by Anurag Dubey from Manifest Global.