this deck is not a drawing. it rebuilds itself from live repository telemetry every 24h. sync stamp is on the panel.
root@localhost:~# ./initialize --operator jesun
[ OK ] identity ................. Jesun Ahmad Ushno
[ OK ] role ..................... AI Engineer & Data Architect
[ OK ] station .................. Toronto, ON
[ OK ] egress policy ............ DENY ALL
[ OK ] integrity ................ SHA-256 verified
[ WARN ] cloud dependency ......... none detected. proceeding.
root@localhost:~# cat /etc/doctrine
[1] The cheapest way to secure data is to never collect it.
[2] Compliance is a design constraint, not a document.
[3] If it can run in the tab, it has no business on a server.
[4] A rule worth having is a rule the code enforces, not one a doc describes.
[5] Ship the system, never the mockup.Four years of IT advisory and ISO 27001 audit work at KPMG taught the same lesson on every engagement: the breach you cannot have is the one on data you never held. So I stopped designing systems that collect, and started designing systems that compute where the data already lives.
Then the same instinct went up a level. A single agent doing the work is a system that collects trust. A fleet with explicit autonomy tiers, adversarial verification and an owner gate is a system that earns it.
Work does not go to one agent. It goes through a lattice. Intake scopes and gates it, plan fixes acceptance criteria before a line is written, worker agents run in parallel on isolated git worktrees, and nothing reaches ship without an adversarial verification pass and an explicit owner signoff.
Autonomy is tiered, not granted. Every seat in the fleet sits in one of three bands, and the band decides what it may do unsupervised:
| tier | authority | gate |
|---|---|---|
AUTO |
executes and commits without asking | post-hoc audit trail |
SIGNOFF |
executes, holds at the boundary | human approves before merge |
ANCHORED |
may not act, may only propose | owner authorises every step |
flowchart TB
OWNER(["OWNER GATE"]):::owner
ORCH{{"ORCHESTRATOR"}}:::core
OWNER --> ORCH
subgraph DIV["DIVISIONS"]
direction LR
D1["Build"]:::seat
D2["Verify"]:::seat
D3["Audit"]:::seat
D4["Research"]:::seat
end
ORCH --> D1 & D2 & D3 & D4
D1 --> W1["worktree/build-*"]:::pool
D2 --> W2["worktree/verify-*"]:::pool
D3 --> W3["worktree/audit-*"]:::pool
D4 --> W4["worktree/research-*"]:::pool
W1 & W2 & W3 & W4 --> LEDGER[("DECISION LEDGER<br/>SHA-256 anchored")]:::ledger
LEDGER -.->|"fail → replan"| ORCH
LEDGER ==>|"pass"| SHIP(["SHIP"]):::ship
classDef owner fill:#1c1408,stroke:#f59e0b,color:#fbbf24,stroke-width:2px
classDef core fill:#022c34,stroke:#67e8f9,color:#a5f3fc,stroke-width:2px
classDef seat fill:#020609,stroke:#2dd4bf,color:#5eead4
classDef pool fill:#020609,stroke:#0e7490,color:#0e7490
classDef ledger fill:#022c34,stroke:#22d3ee,color:#a5f3fc
classDef ship fill:#022c1a,stroke:#4ade80,color:#4ade80,stroke-width:2px
Every decision is logged before it is acted on, not after. The ledger is the system of record, so a session that starts cold reads state and continues rather than guessing.
The other half of the doctrine. Data does not travel to the compute. The compute travels to the data.
flowchart LR
subgraph M["YOUR MACHINE · TRUST BOUNDARY"]
direction TB
A["Raw data"] --> B{{"WASM + Pyodide runtime"}}
B --> C["Transcode · Analyse · Encrypt"]
C --> D["Result"]
end
B -. "blocked" .-> S[("Vendor cloud")]
D -- "WebRTC P2P" --> P["Trusted peer"]
classDef safe fill:#022c34,stroke:#22d3ee,color:#5eead4,stroke-width:1px
classDef danger fill:#2b0a0a,stroke:#ef4444,color:#fca5a5,stroke-width:1px
class A,B,C,D,P safe
class S danger
| system | payload | runtime | |
|---|---|---|---|
01 |
PRISM | Zero-trust analytics. A full Python statistics engine executing inside the tab, nothing transmitted. | TypeScript Pyodide Web Workers |
02 |
BigFish | Autonomous betting terminal. A swarm orchestrator dispatching tactical, sentiment and execution agents against EV thresholds with Kelly sizing. | JavaScript Multi-agent |
03 |
NoPara | Serverless media conversion. Transcoding runs client-side, so the file never leaves the machine. GDPR and PIPEDA by construction. | FFmpeg.wasm PWA |
04 |
TriDrop | iPhone to Windows transfer with no intermediary host. Scan, connect, stream peer to peer. | WebRTC STUN/TURN |
05 |
dingdong-bms | Building management across tenants, maintenance and billing. | Next.js React TypeScript |
06 |
SonicClear | Speech restoration pipeline. Studio-grade output from noisy input. | Flask NumPy SciPy |
07 |
drawing_modal | Canvas capture instrument recording 20+ measurements per point, including velocity, twist and in-air hover trajectory. | Canvas Serverless |
Active build. Prime OS, a multi-tier agent operating layer with a 191-seat org chart, autonomy banding and structural policy enforcement. Not a documentation exercise: a violated rule raises, it does not warn.
| domain | instruments |
|---|---|
| Orchestration | Multi-agent fleets · autonomy tiering · git worktree isolation · adversarial verification · decision ledgers |
| Edge compute | WebAssembly · FFmpeg.wasm · Pyodide · Web Workers · Service Workers · WebCrypto |
| Realtime | WebRTC DataChannels · STUN/TURN · WebSockets · Node signaling |
| Data | pandas · NumPy · SciPy · scikit-learn · TensorFlow · Power BI · Tableau |
| Assurance | ISO/IEC 27001:2022 · GDPR Art. 32 · PIPEDA · OWASP Top 10 · WCAG 2.2 AAA · SHA-256 integrity anchoring |
| Method | Lean Six Sigma · DMAIC · phase-gated execution · acceptance criteria before code |
▚ RESEARCH LOG
Two peer-reviewed conference papers, AIP Conference Proceedings (2023):
- Fabric-defect detection using wavelet transform with neural networks
- An elitist variant of particle swarm optimization
Indexed at ORCID and Google Scholar.
Writing. Beyond the Firewall, arguing cybersecurity should be run as a measurable DMAIC process rather than a posture. Plus a beginner's guide to data analytics (ResearchGate, 2025).
▚ SERVICE RECORD
| period | post |
|---|---|
| 2025 to present | Gallea Ai · AI Engineer & Data Architect |
| 2023 to 2024 | KPMG · IT Advisory, Officer to Senior Officer to Consultant. 20+ Power BI dashboards, Power Automate workflows saving 200+ hours, ISO 27001 audits across banking and healthcare. |
| 2021 to 2023 | ESAB, AIUB · Student Engagement Facilitator, 300+ students |
Training. MSc Data Analytics, University of Niagara Falls Canada (2025 to 2026), holding the UNF Academic Achievement Entrance Award and the Academic Performance Scholarship. BSc Computer Engineering, American International University-Bangladesh.
Certified. Oracle Cloud Infrastructure AI Foundations · Certified Network Security Practitioner (CNSP) · Cisco CyberOps Associate, Network Defense, Endpoint Security, Cyber Threat Management · IBM Data Fundamentals · Lean Six Sigma White Belt
▚ FIELD OPERATIONS
Engineered donation-tracking frameworks and coordinated humanitarian logistics networks for winter asset allocation and flood relief, reaching 50,000+ regional families. Same systems engineering, different problem domain.
session terminated · no data written to remote