# Hanzo — chat, or build with dev

> Ask a question, or build what you describe. The whole cloud is one binary and one API — run it here, on your own Kubernetes, or on your laptop. Open source.

[Models](https://hanzo.ai/models)/Agentic Coding (ENG-401)

Course ENG-401 · Credential: HACE

# Agentic Coding Systems with Hanzo Dev & Zen 6

Learn to architect production-grade coding agents that can inspect repositories, reproduce failing tests, generate syntax-safe diffs, and self-heal in isolated gVisor sandboxes using Zen 6 and ZAP zero-copy RPC.

[Enroll in ENG-401 — $199 USD (+$50 Credit Rebate)](https://hanzo.ai/login?redirect=/university/agentic-coding)[Back to University Directory](https://hanzo.ai/university)

SPECIFICATIONSDURATION4 Weeks

EFFORT8–10 hrs / week

PREREQUISITEPython or Rust / Git

GRADINGSWE-bench Grader

All coursework executes in live Hanzo Dev sandboxes running Zen 6 inference models, tree-sitter AST parsers, and automated CI test runners.

TUITION & REBATE$199USD

Usage Credit Rebate:+$50 USD

25% rounded up deposited immediately into your Hanzo Cloud balance to fund GPU pods and sandbox leases.

[Enroll Now in ENG-401 →](https://hanzo.ai/login?redirect=/university/agentic-coding)Includes automated CI grading & W3C Verifiable Credential

Curriculum Roadmap

## Weekly Module Breakdown

Every week combines architectural theory, repository deep-dives, hands-on terminal labs, and automated evaluation milestones.

Week 1Module 1

Runtime Boundaries & Local Model ServingSet up local high-throughput serving for Zen 6 (27.3B) and Zen 6 Flash. Build gVisor pod sandboxing with lease lifecycles and ZAP zero-copy RPC channels.Zen 6 architecture: quantized GGUF/AWQ serving on Apple Silicon and Linux CUDA

The 4-surface boundary: CLI, SDKs, HTTP REST, and MCP sockets

ZAP Zero-Copy RPC over Cap’n Proto: eliminating JSON serialization latency

gVisor / runsc container virtualization and ephemeral root filesystem mounts

Lab 1: Provision an isolated multi-process execution jail with ZAP IPC socket bridge.

Week 2Module 2

Repository Exploration & Context DistillationTrain agents to navigate 100k+ line codebases without blowing context windows. Combine AST symbol graphs with ripgrep and hybrid semantic search.AST-based symbol extraction using tree-sitter across Python, TypeScript, and Rust

Dual-index code retrieval: BM25 keyword search plus dense vector embeddings

Context compression: distilling 50-file call stacks into 4k-token agent scratchpads

Memory tiers: working scratchpad, thread history, and long-term vector stores

Lab 2: Implement a semantic file scout that isolates bug locations in < 3 tool hops.

Week 3Module 3

Test-Driven Self-Healing & Kai VerificationClose the feedback loop. Have the agent parse issue descriptions, reproduce errors with minimal reproduction test cases, and evaluate patch progress with Kai.Automated reproduction generation: writing failing pytest / cargo test cases from issues

AST diff validation: ensuring patches do not introduce syntax or type regressions

Kai decision model checkpoints: zero-token confidence and verification scoring

Micro-USD budgeting: halting runaway iteration loops when budget caps are reached

Lab 3: Build an autonomous test runner that iterates on patches until exit code 0.

Week 4Module 4

SWE-bench Capstone & Swarm OrchestrationDeploy multi-agent coordinator-executor pairs to resolve authentic GitHub issues from SWE-bench Lite. Submit for automated verification and credentialing.Coordinator-Worker topology: separating strategy planning from file modification

Automated git branch management, commit hygiene, and pull request generation

Telemetry and audit logs: recording full agent trajectories to O11y ClickHouse

Cryptographic credential signing: W3C Verifiable Credentials via Hanzo Trust

Capstone Project: Solve 5 SWE-bench benchmark issues with 100% test pass rate and < $0.50 spend per issue.

Automated Evaluation

## SWE-bench Capstone Evaluation Rubric

To earn your Hanzo Certified Agentic Systems Engineer (HACE) credential, your capstone submission is graded against five rigorous automated criteria.

1. Test Suite Pass Rate (40%)All pre-existing unit tests plus student-generated bug reproduction tests must pass with exit code 0.

2. AST & Syntactic Cleanliness (20%)Patches must pass strict linter, formatter, and typechecker without introducing unused imports or style warnings.

3. Micro-USD Budget Adherence (20%)The agent must resolve each problem set within a strict $0.50 (500,000 micro-USD) inference ceiling.

4. Telemetry & Audit Fidelity (20%)Complete Cap’n Proto trace records and Kai decision probabilities must be emitted to Hanzo O11y.

Enroll in ENG-401

## Ready to build autonomous coding systems?

Get instant sandbox access, full syllabus modules, automated SWE-bench test harness, and $50 in Hanzo Cloud usage credits to fund your lab compute.

[Enroll Now for $199 USD (+$50 Credit Rebate)](https://hanzo.ai/login?redirect=/university/agentic-coding)[Explore All Programs](https://hanzo.ai/university)