TEKNIS · BUKTI ENGINEERING

Pusat Bukti Teknis

Bagian ini membuktikan apa yang baru saja terlihat pada demo Bambang. Alurnya sama: panggilan masuk → kasus dibuat → bukti ditata → Qwen3.7-Plus menganalisis keluhan dana Rp128,75 juta yang belum kembali → manusia meninjau → jejak audit tercatat.

← TAMPILAN EKSEKUTIF / BISNIS

BUKTI TEKNIS SCENARIO A

Kasus teknis yang sama dengan demo Bambang

Tidak ada cerita teknis yang terpisah. Semua modul di bawah menjelaskan implementasi di balik perjalanan Bambang: intake keluhan, case, evidence, analisis AI, human review, workflow, automation, security, dan audit.

KONTEN TEKNIS

Simulator Skenario

LAB UJI

Jalankan demo proses/simulasi dan panggil Qwen3.7-Plus secara live pada tahap analisis. Hasil AI tetap berhenti pada tinjauan manusia.

OPEN TECHNICAL TEST →

Operator Workspace

LIVE

Periksa case, evidence, AI recommendation, maker-checker actions, dan audit timeline dari sudut pandang operator.

OPEN TECHNICAL TEST →

Runtime & Capability

What is actually working

TERVERIFIKASI PROTOTYPE

Prototype B Web

LIVE

Isolated preview environment behind nginx; application port remains loopback-only.

Runtime AI Live

TERVERIFIKASI

Qwen3.7-Plus menjalankan analisis live pada kasus Bambang: dana Rp128,75 juta yang belum dikembalikan.

Tata Kelola Manusia

PASS

Hasil AI live berhenti pada HUMAN_MAKER_REVIEW; model tidak memiliki kewenangan persetujuan.

Penyimpanan Data

AKTIF

PostgreSQL + PostgREST persistence for Scenario A operational data.

Alur Event

AKTIF

Redis stream, outbox worker and event bridge are running.

Otomasi

AKTIF

n8n orchestrator requests allowlisted actions through service identity.

Analisis AI live yang terverifikasi

Bukti analisis Qwen pada kasus Bambang

Model
Qwen3.7-Plus
Live test latency
31 seconds
Execution status
COMPLETED
Confidence
0.30
Recommendation
manual-review
Tinjauan manusia
Wajib
Tindakan berikutnya
HUMAN_MAKER_REVIEW
Hasil tata kelola: AI menyusun analisis kasus, tetapi tidak diberi kewenangan untuk mengambil tindakan penting secara mandiri. Model merekomendasikan tinjauan manual dan alur berhenti pada kewenangan manusia.

Engineering modules

Inspect implementation and controls

Arsitektur

TERIMPLEMENTASI

Layer ownership, component boundaries, Scenario A overlay, service connections and control points.

OPEN EVIDENCE →

Basis Data / ERD

TERIMPLEMENTASI

Persistent domain model, entity ownership, relationships, system-of-record map and Scenario A data path.

OPEN EVIDENCE →

API & Events

TERIMPLEMENTASI

REST contracts, domain events, producer/consumer ownership, permissions and non-custodial boundaries.

OPEN EVIDENCE →

Alur Kerja

TERIMPLEMENTASI

Controlled state machine, human/AI/service tasks, maker-checker gates and invalid-transition protection.

OPEN EVIDENCE →

AI Runtime

LIVE TERVERIFIKASI

Qwen3.7-Plus model boundary, tool allowlists, execution limits, structured output and escalation.

OPEN EVIDENCE →

Keamanan

TERVERIFIKASI

Distinct human/agent/service identities, RBAC, segregation of duties, data classification and fail-closed policies.

OPEN EVIDENCE →

Otomasi / n8n

LIVE

Redis event bridge, n8n orchestration boundary, automation allowlist, service identity and idempotency controls.

OPEN EVIDENCE →

Infrastruktur

LIVE

Prototype B topology with PostgreSQL, Redis, PostgREST, workers, n8n, nginx and loopback-only application exposure.

OPEN EVIDENCE →

Known limits / Next hardening

What we are not claiming yet

  • Prototype B is a governed working prototype, not yet a production deployment.
  • Live LLM scope is intentionally restricted to the synthetic Scenario A resources.
  • Current live-model latency is approximately 30 seconds and should be optimized before production scale.
  • External production integrations, production IAM and production observability remain future hardening work.
Non-negotiable boundary: AI has no material approval authority; n8n is not system of record; TP2A remains non-custodial; business-state mutations pass through owning services and create audit evidence.