EnterpriseHub Hiring Review Guide

EnterpriseHub is best reviewed as an AI backend engineering portfolio project: a real estate lead-qualification platform with FastAPI APIs, multi-bot orchestration, LLM caching, compliance checks, CRM sync, evals, and observability-oriented infrastructure.

This guide is intentionally short. It points reviewers to the strongest evidence first and separates proven artifacts from roadmap items.

60-Second Read

• Problem: real estate teams need fast, compliant lead qualification across inbound SMS/web leads.
• Backend system: FastAPI services coordinate lead, buyer, and seller bot workflows with CRM persistence and webhook handling.
• AI system: prompt registry, golden dataset, deterministic checks, LLM-as-judge harness, adversarial tests, and nightly eval workflow.
• Production judgment: ADRs, security scanning, structured logging, health checks, Docker/Compose deploy paths, Redis/Postgres cache design, and CI gates.
• Current caveat: public claims need stricter provenance. Some README/benchmark/case-study metrics mix measured results, design targets, and projections.

5-Minute Review Path

1. Read the top of README.md through "For Hiring Managers".
2. Read CASE_STUDY.md, especially "Honest Production Metrics".
3. Inspect the eval surface in evals/README.md, evals/judge.py, and tests/test_eval_harness.py.
4. Skim the architecture decisions in docs/adr.
5. Read the hiring audit findings in docs/HIRING_CONVERSION_AUDIT.md.

30-Minute Technical Review Path

1. LLM orchestration: ghl_real_estate_ai/services/claude_orchestrator.py
2. Agent routing and governance: ghl_real_estate_ai/services/agent_mesh_coordinator.py
3. Webhook and CRM boundary: ghl_real_estate_ai/api/routes/webhook.py
4. Compliance pipeline: ghl_real_estate_ai/services/jorge/response_pipeline/pipeline.py
5. Eval harness: evals/judge.py, evals/golden_dataset.json, evals/baseline.json

Local Verification Commands

These commands were audited on April 29, 2026. Some fail today; that is useful signal for the next development phase.

ruff check .
ruff format --check .
mypy ghl_real_estate_ai src utils advanced_rag_system
pytest --collect-only --override-ini='addopts='
pytest tests/test_eval_harness.py --override-ini='addopts=' -q
pytest tests/unit/test_claude_orchestrator.py tests/unit/test_sql_safety.py --override-ini='addopts=' -q
pytest tests/api/test_health_routes.py --override-ini='addopts=' -q
pytest tests/security/test_webhook_signatures.py --override-ini='addopts=' -q

Known Review Caveats

• Global lint and format checks currently fail because of parse errors and formatting drift, concentrated heavily in advanced_rag_system.
• Full-repo mypy did not complete locally during the audit; create a bounded type-check command for the flagship API/services.
• pytest --collect-only --override-ini='addopts=' currently collects 7,721 tests with 38 skipped; public test-count claims should use the current reproducible count.
• FastAPI route metadata is uneven: an AST scan found 702 route decorators, 427 without response_model, and 677 without explicit status_code.
• Some security/health targeted tests fail locally, indicating either test drift, route drift, or environment assumptions that need tightening.
• The strongest proof is not "big repo size"; it is the combination of orchestration, compliance, eval discipline, and honest production tradeoffs.