Skip to content
§00 // architectural systems & applied AI

Yofal
Rifanda

Senior Systems Architect & AI Automation Creator. I design and build institutional platforms, deterministic PostgreSQL backends and applied-AI pipelines that hold up under real-world load.

Available for contractUTC+8 (WITA)
capability matrix
rig-01 · dossier monitor
live
uptime
99.98%
p95 latency
42ms
deploys / wk
18
incidents
0
packet stream
workstation
host
rig-01 · wita
cpu
ryzen 9 · 16c/32t
memory
64 gb ecc
display
dual 27" · 4k
editor
neovim + tmux
shell
zsh · ghostty
0°48'N 117°29'E
--:--:-- WITA
§01selected work

Systems shipped to production.

Institutional platforms and applied-AI pipelines, each with its live interface and the architecture beneath it. Toggle any case between the rendered UI and its system blueprint.

01

SIMPEL&PAS

Integrated civil-service permit & licensing platform

Public sector · LicensingLead systems architect · 2024
client scope

A single-window permit engine replacing 14 paper workflows across 9 regional offices, with a deterministic approval state machine and full audit trail.

UAT verification
  • UAT cases passed312 / 312
  • Rollback events0
  • Office rollout9 / 9
latency & load benchmarks
Approval p952.1s
target < 3s
Permits / month18,400
sustained
Error budget0.02%
30d window
Next.jsPostgreSQLRedisS3RBAC
app.simpelpas.io
approved
submit
review
verify
issue
PRM-8841issued
PRM-8842in review
PRM-8843verifying
PRM-8844queued
9 offices · livep95 2.1s
erd · system blueprintrev.2024Citizen PortalSSR · edgeOfficer ConsoleRBACAPI Gatewayrate · authzPermit FSMstate machinePostgreSQLRLS · auditJob Queueredis streams
Provincial Government · One-Door Servicecase 01 / 05
02

BIZCONTROL

Real-time operations & finance control tower

Logistics · FinOpsBackend & data architect · 2024
client scope

A streaming control tower unifying fleet telemetry, settlements and margin analytics into one deterministic ledger with sub-second dashboards.

UAT verification
  • Reconciliation drift0.00%
  • Load test peak9.4k rps
  • Data lossnone
latency & load benchmarks
Ingest p9938ms
target < 50ms
Events / day42M
streaming
Ledger lag0.4s
end to end
TypeScriptPostgreSQLKafkaClickHousegRPC
app.bizcontrol.io
42M
events
38ms
p99
0.4s
lag
throughput · 60s
erd · system blueprintrev.2024Fleet EdgetelemetryOps Cockpitlive UIStream BuskafkaLedger Coreexactly-onceClickHouseOLAPPostgreSQLOLTP
Regional Logistics Groupcase 02 / 05
03

WARGAHUB

Citizen engagement & service-request mesh

Civic tech · CommunityFull-stack architect · 2023
client scope

A geo-aware service-request network routing citizen reports to the right unit with SLA tracking, deduplication and a public transparency feed.

UAT verification
  • SLA compliance97.3%
  • Duplicate rate1.1%
  • Districts live22 / 22
latency & load benchmarks
Route time0.9s
report to unit
Reports / month26,700
active
Resolution91%
within SLA
Next.jsPostgreSQLPostGISWorkersWebSockets
app.wargahub.io
22 districts
91%
resolved
0.9s
routing
erd · system blueprintrev.2023Citizen AppPWAAgency DeskrealtimeGeo RouterpostgisSLA EnginetimersPostgreSQLpostgisRealtime Hubws fanout
Municipal Smart-City Officecase 03 / 05
04

NOVA

Autonomous AI operations copilot

Applied AI · AutomationAI systems architect · 2025
client scope

An agentic copilot that plans, executes and verifies multi-step operational workflows against typed tools, with human-in-the-loop checkpoints and full replay.

UAT verification
  • Eval pass rate94.2%
  • Hallucinated calls0.3%
  • Replayable runs100%
latency & load benchmarks
Plan latency1.4s
first token
Tool calls / run11.2
avg
Autonomy78%
no handoff
TypeScriptLLM ToolsVector DBQueuesEvals
app.nova.io
nova · agent run #4821

▸ plan: resolve ticket backlog

01 retrieve(context) ok

02 classify(intent) ok

03 call tool: db.query ok

04 verify(output) · gate

05 draft response ...

eval score 0.94 · autonomy 78%

$run --replay
erd · system blueprintrev.2025Operator UIstreamPlannerLLMOrchestratorgraphTyped ToolsguardedVector StoreretrievalEval Gateverifier
Internal R&D · Enterprise pilotscase 04 / 05
05active pipeline

Next-Gen Autonomous AI Workflows

An R&D track exploring self-verifying agent graphs that plan, execute and audit long-horizon operational work against typed tools, with deterministic replay and human checkpoints. In active development, selected pilots underway.

Agent graphsTyped toolsEval gatesReplayHITL
§02code execution sandbox

Security is a property of the system, not a feature.

A side-by-side of hardened production code against the naive version that ships in a hurry. Run the live trace to watch the same payload breach one path and get stopped by the other.

SQL Injection · CWE-89
payload ' OR 1=1 --

Unsanitised input concatenated into a SQL string lets an attacker rewrite the query and dump every row.

Naive · Vulnerable
api/login.ts
1// Builds the query by gluing input into a string2async function findUser(email: string) {3  const sql =4    "SELECT * FROM users WHERE email = '" + email + "'"5  return db.query(sql)6}
Hardened · Production
api/login.ts
1// Parameters are bound, never interpolated2async function findUser(email: string) {3  const sql =4    "SELECT * FROM users WHERE email = $1"5  return db.query(sql, [email])6}
naive trace

$ awaiting trace — press run live trace

hardened trace

$ awaiting trace — press run live trace

crypto telemetry
seq 0000
sha-256 · integrity
computing…
hmac tag

signing…

nonce…
principles
  • Never trust input · bind, never concatenate
  • Secrets hashed, compared in constant time
  • Authorize every read against the caller
  • Fail closed · log · surface nothing sensitive