Design from first principles
Choose abstractions because they fit the problem—not because a framework made them convenient.
September cohortEarly-bird saves ₦100,000 until 21 AugustClaim your seat →
Applications open · Cohort 04
Stop learning frameworks.
Learn the architecture.
Start with nothing but an LLM. In 12 intensive weeks, build the tools, memory, workflows, safety systems, and operational discipline behind production-grade AI agents.
12 live weeks7 production labs6 frameworks compared
One mental model.
Every serious agent stack.
01 / THE PROBLEM
Most agent courses teach APIs that expire.
Framework tutorials create framework operators. This bootcamp creates agent engineers.
You will understand the durable mechanics beneath the libraries: state, tool contracts, orchestration, identity, approvals, recovery, evaluation, and governance.
When the framework changes, your mental model still works.
02 / THE TRANSFORMATION
Build the system
one capability at a time.
Each stage solves a real engineering failure in the system you built the week before.
01MODEL
02TOOLS
03MEMORY
04WORKFLOWS
05TRUST
06PRODUCTION
Ship vertical slices
Implement typed tools, persistent state, durable workflows, evaluation gates, and observability.
Prove production readiness
Demonstrate permissions, recovery, cost controls, human oversight, and safe failure behavior.
03 / 12-WEEK CURRICULUM
The production agent stack.
Weekly releases keep the cohort together. Every lesson ends in evidence: a working lab, an architecture decision, or an evaluated system.
The model is not the agent
Tokens, context, inference, uncertainty, structured outputs, and the boundary between intelligence and agency.
YOU SHIPA defensible model card and typed response contract.
The smallest useful agent loop
Observe, decide, act, evaluate, repeat. Add termination, budgets, and evidence to a controlled runtime.
YOU SHIPA deterministic agent loop with traceable state.
Give the model hands—safely
Tool schemas, policy checks, side effects, idempotency, timeouts, retries, and uncertain outcomes.
YOU SHIPA permissioned tool registry and risk-classified action.
Remember with purpose
Working, episodic, semantic, and procedural memory; retention, consent, correction, and provenance.
YOU SHIPA scoped memory service with lifecycle controls.
Ground decisions in evidence
Retrieval, chunking, ranking, citations, access control, and evaluation of answer groundedness.
YOU SHIPA cited RAG pipeline with retrieval tests.
Reason under constraints
Plan representation, dynamic replanning, acceptance tests, search budgets, and non-progress detection.
YOU SHIPA planner that exposes decisions—not hidden thought.
Make execution durable
State machines, graphs, checkpoints, replay, parallel branches, compensation, and versioning.
YOU SHIPA resumable workflow that survives interruption.
Put people in the loop
Approval packets, binding scope, escalation, expiry, separation of duties, and safe resume.
YOU SHIPA review queue with cryptographically bound approvals.
Coordinate specialized agents
Handoffs, agents-as-tools, shared state, isolation, termination, and the single-agent baseline.
YOU SHIPA bounded multi-agent system with a cost case.
Secure the whole path
Identity, least privilege, prompt injection, confused deputy attacks, guardrails, and audit boundaries.
YOU SHIPA threat model and layered control plan.
Measure, recover, improve
Traces, metrics, evals, red teams, SLOs, incident response, cost controls, rollout, and rollback.
YOU SHIPA release gate backed by trajectory evaluations.
Defend your architecture
Vertical integration, production-readiness review, capstone demonstration, and oral defense.
YOU SHIPA deployable agent and production-readiness dossier.
04 / FRAMEWORK FLUENCY
Six dialects.
One architecture.
Every core idea is translated across the ecosystem so you can evaluate tools, join unfamiliar codebases, and avoid framework lock-in.
STACKBEST LENSWHAT YOU MASTER
OpenAI Agents SDKMinimal primitives
Agents, tools, handoffs, guardrails, sessions, and tracing.
LangGraphDurable state graphs
Nodes, edges, checkpoints, interrupts, and replayable workflows.
CrewAIRoles and flows
Agents, tasks, crews, processes, and event-driven flows.
AutoGenConversation systems
AgentChat teams, messages, tools, termination, and runtimes.
Google ADKComposable agent apps
Agents, runners, sessions, artifacts, workflows, and deployment.
OpenClawPersistent gateway
Channels, skills, schedules, workspace policy, and operations.
05 / COHORT EXPERIENCE
Built to be completed.
This is a live operating cadence—not a content library. Deadlines, feedback, peer accountability, and visible progress turn intention into finished work.
MON
Module release
Lectures, annotated examples, reading, quiz
WED
Live architecture studio
Concept briefing, teardown, worked example
FRI
Assignment checkpoint
Submit evidence and receive structured feedback
SAT
Build clinic
Debug live systems with instructors and peers
LEARNER WORKSPACE
GOOD MORNING, AMINA
Your agent is becoming durable.
63%complete
NEXT DEADLINE · FRIDAYWorkflow recovery lab
3 of 4 acceptance checks passed
06 / GRADUATION STANDARD
You leave with proof,
not participation.
Your capstone is a production-readiness dossier: an end-to-end agent, architecture records, threat model, evaluation suite, telemetry, runbook, and live defense.
HYRAX AI ACADEMY
CERTIFICATE OF COMPLETION
Agentic AI
from First Principles
Production Agent Engineer · Cohort 04
H
07 / SIGNALS
A curriculum designed
to change the conversation.
Launch note: replace these positioning previews with verified graduate testimonials after the pilot cohort.
“The missing layer between prompt engineering and real AI systems engineering.”
“It makes every framework feel like a dialect of the same underlying language.”
“The capstone asks the question most courses avoid: would you actually operate this?”
08 / ENROLLMENT
One cohort.
A career-long advantage.
For engineers, product leaders, founders, consultants, and public-sector digital teams ready to build dependable agent systems.
- 36+ hours of live teaching and build clinics
- Lifetime access to the course edition and updates
- Code labs, architecture templates, and assessment tools
- Private cohort community and instructor feedback
- Verified certificate after capstone defense
EARLY-BIRD · SAVE ₦100,000
INDIVIDUAL SEAT₦450,000
Pay once. Two-payment plan available on request.
Apply & secure your seatSecure Flutterwave-hosted checkout · Receipt issued automatically
Enrolling a team?
Private cohorts and team pricing are available for five or more participants.
Request a team proposal →COHORT 0414 Sep — 4 Dec 2026Applications open · 18 seatsApply now →
COHORT 0525 Jan — 16 Apr 2027Priority list openJoin priority list →
PRIVATE TEAM COHORTDates tailored to your team5–30 participantsRequest proposal →
09 / FAQ
Before you apply.
Is this course for beginners?+
You should be comfortable reading and editing Python. We teach agent engineering from first principles, but this is an intensive professional program rather than an introduction to programming.
Do I need prior experience with every framework?+
No. The course deliberately begins without a framework. You learn the universal mechanism first, then translate it into six major ecosystems.
What happens if I miss a live session?+
Recordings and materials are posted within 24 hours. Live attendance is strongly encouraged, and every learner receives one make-up clinic during the cohort.
How is the certificate earned?+
Complete at least 80% of graded work, pass the core safety and reliability gates, submit the capstone dossier, and defend the final system in a live review.
Are payment plans or team seats available?+
Yes. A two-payment plan is available by request. Teams of five or more can book a private cohort with tailored case studies and reporting.
Will the course stay useful as tools change?+
Yes—that is the point. APIs are treated as implementations of enduring concepts: state, tools, workflows, permissions, evaluation, and operations.
APPLICATIONS CLOSE 10 SEPTEMBERBuild agents you would
Join the September cohort
Build agents you would
trust in production.
Join the September cohort18 seats remaining · Admission reviewed within 48 hours