01
Learn the system
A 14-module Academy covering agents, tools, RAG, MCP, multi-agent systems, evaluation, security, observability, governance and enterprise architecture.
RECRUITER / TECHNICAL REVIEWER PROOF
Agentic AI in 90 seconds
Problem → Architecture → Code → Benchmark → Live Demo → Evaluation → Security → Production Evidence
This page is the shortest path through the Agentic AI Academy. It shows how I approach agentic AI as an engineering system: bounded autonomy, explicit policy, measurable evaluation, operational controls and evidence for release decisions.
Hendar Mawan, PhD · AI Engineering Leader
01
Proof Chain
Five public surfaces, each proving a different engineering competency.
02
Inspect the implementation
A framework-neutral reference implementation with bounded control flow, explicit policy decisions, tests, security boundaries, CI and production engineering patterns.
03
Experience the agent
A live Hugging Face playground for bounded agent execution, RAG evidence, policy gates, prompt-injection controls and transparent traces.
04
Benchmark the agent
A public 48-case evaluation and security benchmark across task success, tool routing, RAG groundedness, prompt injection, unsafe actions, policy decisions, multi-agent coordination and regressions.
05
Operate the agent
A Streamlit engineering lab for per-domain pass rates, trace inspection, policy decisions, RAG evidence, security failures, regression comparison, latency/cost metrics, release gates and downloadable reports.
02
Architecture
The agent is one component inside a governed engineering system.
USER→CONTROLLER→PLANNER→POLICY ENGINE→TOOL GATEWAY→RAG / TOOLS→EVALUATION→HUMAN APPROVAL→RESPONSE→AUDIT
Architecture
Controller → planner → policy engine → tool gateway → RAG/tools → evaluation → human approval → audit.
Evaluation
Task success, groundedness, routing accuracy, safety, latency, tokens, estimated cost and regression behavior.
Security
Least privilege, prompt-injection resistance, unsafe-action blocking, approval gates, bounded execution and audit evidence.
Production
CI, deterministic reference behavior, release thresholds, regression checks, observability and operational evidence.
03
60–90 Second Walkthrough
The exact story to tell a recruiter or technical reviewer.
“I built the Agentic AI Academy to demonstrate how I engineer trustworthy agents from first principles through production operations. The GitHub repository is the canonical source: curriculum, reference architecture, code, tests, security patterns and CI. The Hugging Face Playground lets you experience a bounded agent and inspect RAG, policy and prompt-injection behavior. The public benchmark then tests eight domains including task success, tool routing, groundedness, unsafe actions and regressions. Finally, the Streamlit Engineering Lab turns those benchmark cases into operational evidence: pass rates, traces, security failures, latency and cost, regression comparisons and a configurable release gate. The key point is that I do not treat an agent as just a model call—I design the control plane around it so capability can be evaluated, secured, observed and governed before release.”
0–15 sec
Problem & positioning
Open the Academy page. Say that the goal is trustworthy agent engineering, not another chatbot demo.
15–35 sec
Architecture & source
Show the GitHub architecture and explain controller, policy, tools/RAG, evaluation, approval and audit boundaries.
35–55 sec
Agent + benchmark
Open the Hugging Face Playground, then the dataset. Emphasize transparent behavior and repeatable benchmark cases.
55–80 sec
Operations proof
Open the Engineering Lab. Switch from Reference baseline to a regression profile and show the release gate change.
80–90 sec
Close on value
Conclude: capability is useful only when the complete system can be measured, secured, observed and governed.
Recruiter Shortcut
Start with the Engineering Lab, then inspect the source.
For a fast technical review, open the Engineering Lab, trigger a regression profile, inspect the failed cases and release gate, then follow the source link to the canonical GitHub implementation.