Lead Instructor: Agentic AI Engineering Program

About Edversity

Edversity is a technology learning platform based at the National Science & Technology Park, NUST, Islamabad. We have trained more than 12,000 learners through live programs in Artificial Intelligence, Cybersecurity, Cloud, Blockchain, Data, and emerging technologies.
We are launching our Agentic AI Engineering Program, a hands-on specialization that prepares experienced developers to build, evaluate, and ship production AI agent systems.
We are looking for an experienced industry practitioner to lead the program.

Who We Are Looking For

We are not looking for someone who has published prompt-engineering content, completed an AI course, or built demos.
We are looking for a practitioner who has put an LLM system in front of real users, watched it fail in ways nobody predicted, and fixed it. Someone who has written an evaluation suite, argued with a client about running costs, designed tools that a model kept misusing until the descriptions were rewritten, and knows the difference between an agent that demos well and one that survives a Monday morning.
The right instructor should be able to help a student move from:
Backend / Full-stack Developer → AI Engineer → Agent Systems Engineer → AI Platform Engineer

The Role

You will lead a 10-week live program designed to prepare learners for roles including:

• AI Engineer
• Agent Systems Engineer
• LLM Application Engineer
• AI Platform Engineer
• Forward-Deployed AI Engineer
• Independent AI automation consultant

This is a highly practical teaching role.
You will deliver live technical sessions, review student code and evaluation suites, mentor capstone projects matched with real business stakeholders, and help assess students for internships and employer referrals.
The program is structured as eight teaching weeks followed by two build weeks. In Weeks 9 and 10 there are no lectures — you move from instructor to technical consultant while students ship.
The role is designed to run alongside a full-time engineering position.

Key Responsibilities

• Deliver two live technical sessions each week across the eight teaching weeks
• Teach through live coding, debugging, and deliberate demonstration of failure
• Guide students in building agent systems from the loop upward, without relying on framework abstractions
• Review student code, GitHub repositories, evaluation suites, and weekly project submissions against published acceptance criteria
• Provide written technical feedback within 48 hours of each submission
• Run design reviews, one-to-one technical sessions, and open office hours during the build weeks
• Mentor students through a capstone deployed for a real business stakeholder
• Help students diagnose silent failures, retrieval quality problems, cost overruns, and guardrail gaps
• Prepare the course's failure material: broken tool sets, injectable bugs, and adversarial prompt sets used in live sessions
• Teach students to scope, price, and hand over agent projects as paid work
• Keep program content aligned with current agent engineering practice as SDKs and tooling change
• Identify high-performing graduates suitable for internships and employer referrals
• Collaborate with Edversity's teaching assistants, cohort team, and industry reviewers

Program Areas

The program covers:
Agent Loop Engineering

• Claude model family and cost economics
• Messages API and content blocks
• Tool use and stop-reason control
• Building an agent loop from scratch without an SDK
• Tool failure handling: exceptions, empty results, timeouts, malformed output
• Token accounting and per-run cost reporting

Tool Design and MCP

• Tool descriptions as interface contracts
• Diagnosing and repairing tool misuse
• Schema design, enums, and tool choice
• Model Context Protocol architecture: hosts, clients, servers
• Building and testing MCP servers
• Scoped configuration and permissions

Retrieval and Knowledge Systems

• Embeddings and semantic search
• Chunking strategies and their effect on retrieval quality
• Vector stores, from SQLite to hosted options
• Hybrid search and reranking
• Building labelled question sets and measuring recall
• Recognising when retrieval is the wrong approach

Structured Output and Document Pipelines

• Schema design with Pydantic
• Designing for absence so systems report "not found" instead of fabricating
• Validators that catch silent numerical correction
• Classify-then-extract pipelines
• Messy real input: scans, multi-page, mixed English and Urdu
• Retry, escalation, and human review routing

Evaluation and Observability

• Testing non-deterministic systems
• Outcome, trajectory, and step-level evaluation
• Golden sets and adversarial test cases
• Model-as-judge techniques and their limits
• Structured trace logging and what must never be logged
• Silent failure detection and regression evaluation

Guardrails and Bounded Autonomy

• Deterministic hooks versus prompt-based rules
• Permission scoping at the argument level
• Human-in-the-loop escalation and approval gates
• Append-only audit trails
• Iteration caps, spend caps, and timeout policies

Multi-Agent Orchestration

• When multi-agent architectures help and when they add cost without benefit
• Orchestrator and subagent patterns
• Scoped context and schema-validated handoff
• Parallel and sequential spawning
• Claude Code, project context files, and reusable skills

Deployment and Operations

• Deploying agent systems and managing secrets
• Rate limiting and infrastructure-level cost caps
• Monitoring: error rate, latency, daily cost, evaluation pass rate
• Handover documentation and runbooks
• Scoping and pricing support retainers

Requirements

• Four or more years of professional software engineering experience, with at least one year building LLM-based systems
• Demonstrable production experience with LLM applications — an agent, pipeline, or RAG system that real users depended on
• Strong Python skills. Comfortable working without framework abstractions
• Hands-on experience with tool use, function calling, and agent loop control
• Practical experience building or consuming MCP servers, or equivalent tool-integration work
• Experience with retrieval systems and the ability to explain why one performs better than another
• Experience writing evaluation suites for non-deterministic systems
• Working knowledge of deployment, secrets management, monitoring, and cost control
• Experience diagnosing production failures in LLM systems: hallucinated fields, silent truncation, cost spikes, degraded retrieval
• Regular use of AI coding assistants such as Claude Code, GitHub Copilot, Cursor, or similar
• Strong communication and mentoring skills
• Ability to explain engineering concepts through practical examples rather than only theory
• Strong communication skills in English and Urdu
• Availability for fixed evening sessions for 10 consecutive weeks

Preferred Experience

Experience in any of the following would be an advantage:

• Ownership of an agent system in production at a product company
• Building evaluation and observability tooling for AI systems
• Cost optimization at scale across model tiers
• Guardrail and policy enforcement for autonomous systems
• Multi-agent orchestration in a real workload
• Document extraction pipelines at volume
• Consulting or freelance delivery of AI systems to SME clients
• Experience with evaluation tooling such as Promptfoo, DeepEval, Ragas, or LangSmith
• Technical interviewing or mentoring junior engineers
• Experience at a software product company or engineering-led technology organization

Previous teaching experience, postgraduate qualifications, and vendor certifications are not required.
We are primarily looking for a strong practitioner who can teach from real engineering experience.

Time Commitment

Approximately 9 to 12 hours per week, including:

• 4 hours of live instruction during the eight teaching weeks
• Session preparation
• Student code and project reviews against published acceptance criteria
• Design reviews and technical clinics
• Capstone mentoring during the build weeks
• Limited student support outside class

Live instruction reduces in Weeks 9 and 10, replaced by one-to-one design reviews, group debugging sessions, and office hours.
Before the first cohort, additional preparation time is required to build the course's failure material: broken tool sets, injectable bugs, and adversarial prompt sets. This is paid separately and scoped with you.
The program runs for 10 consecutive weeks.
You will be supported by:

• Teaching assistants
• A full-time cohort manager
• Practitioner reviewers
• Guest instructors for specialist sessions where required
• Edversity's placement and employer-relations team

What We Offer

• Competitive per-cohort compensation
• Paid curriculum preparation ahead of the first cohort
• Ownership of technical program delivery
• Public visibility as the program's Lead Instructor
• Opportunity to shape one of Edversity's flagship engineering programs
• Access to Edversity's employer and technology ecosystem
• Support from teaching assistants and cohort operations
• Opportunities to lead future cohorts
• Opportunities to participate in employer sessions, Demo Days, and industry events
• A teaching schedule designed to work alongside a full-time engineering role