Delivery leadership · AI solutions architecture
Sixteen years leading delivery across government, banking and telecommunications, May 2010 to September 2026 — currently Scrum Master and Project Manager on the Australian Taxation Office's Payday Super program.
Session observatory
Browser-local sample only. No server, GPU, portfolio or tracking metrics are read or sent.
Melbourne, Australia
self-reported, from my CV. Repository figures below are harvested and dated.
Open to delivery-leadership and AI engagementsLinkedInGitHubEmail
Photograph · Melbourne
About
I build a job-fit engine that scores a candidate against a role on ten dimensions. It is the fairest summary of a person I know how to write, so here I am on all ten of them.
There are no scores below. The engine refuses to publish a number it cannot source, and a number I assign to myself has no source at all — so these are answers, with the evidence beside them.
Dimensions taken verbatim from Victordtesla24/aether-job-career-agentapps/api/app/routers/jobs.py
Equal-radius compass
Checkable source line
Python and TypeScript to production, and REXX on a mainframe when the program actually needs it. Cloud-native across Azure, GCP and AWS; Kubernetes, Terraform, Postgres.
38 public repositories · ATO evidence harness · ANZ platform migrations
Candidate-side answer.
Python and TypeScript to production, and REXX on a mainframe when the program actually needs it. Cloud-native across Azure, GCP and AWS; Kubernetes, Terraform, Postgres.
38 public repositories · ATO evidence harness · ANZ platform migrations
Sixteen years. Senior delivery lead and AI/ML solutions architect since 2017; before that, business analysis and project delivery from May 2010 — sixteen years and four months to September 2026.
ATO · ANZ · NAB · Microsoft · Telstra · InfoCentric · MYOB
Government, banking and telecommunications — three regulated industries where the delivery constraint is usually assurance, not engineering.
Australian Taxation Office, ANZ, NAB, Telstra
Scrum Master and Project Manager who architects. I have never found the two halves to be separate jobs: the eight-squad test-evidence problem at the ATO was a delivery problem solved with a toolchain.
Payday Super program · Agile Kookaburras squad · PI 47–48
Agile as practice rather than ceremony. Cadence and PI planning when the plan holds; a cross-discipline war room, inside three hours, when it does not.
5+ squads, up to 40 practitioners onshore and offshore
Melbourne market band for senior delivery leadership. I would rather agree the scope first and let the rate follow it.
Open to permanent and contract engagements
Melbourne, Victoria. Hybrid locally, or remote across Australian and New Zealand time zones.
Currently on site with the ATO, Melbourne
Programs where AI delivery and AI assurance are the hard part — where somebody has to be accountable for whether the model output can be trusted, not just whether it shipped.
Langfuse + Phoenix evaluation stack · −38% simulated error-budget breaches
I look for organisations that can absorb an honest status report. Every program I have rescued was one where somebody said the number out loud early enough.
75+ hours of evidence against 64 available — escalated, then re-baselined
Build systems whose claims can be checked. Everything I ship is designed to refuse to fabricate its own evidence, and to say so when it cannot measure something.
aether-job-career-agent · unmeasured signals read "not measured", never zero
Experience
Every bar below is drawn to its real duration — the same dates as the CV, on one axis. The long one in the middle is eight years at ANZ, and it is the reason the rest of this reads the way it does.
Three roles state a figure in the CV. The other five state none, and none was invented for them — those carry an open bracket instead.
≈92% (Self-reported figure.)evidence effort removed across 200+ SIT scenarios
−38% (Self-reported figure.)error-budget breaches — measured against a simulated budget, not live traffic
10k+ (Self-reported figure.)concurrent devices at P95 under 200 ms
no published figure (Not measurable; reason given.)
no published figure (Not measurable; reason given.)
no published figure (Not measurable; reason given.)
no published figure (Not measurable; reason given.)
no published figure (Not measurable; reason given.)
Skills & Certifications
Every instrument ships with a certificate saying what was tested, where, and what was not. This is that certificate. There are no proficiency bars on this page, because nobody can check one.
Programmes
Repositories
Credentials
public/docs/Vik_Resume_Final.pdf17 of 17 capabilities shown
| Capability | Evidence | Where | Status |
|---|---|---|---|
| Mainframe test automation — REXX, SMF, SDSF, PCOMM | 200+ SIT/E2E scenarios · ~3 h to ~15 min per scenario | ATO · Payday Super | measured in production |
| Agile delivery at programme scale | 8 squads · PI 47–48 · test capacity re-baselined 30 → 90 person-days | ATO · Payday Super | measured in production |
| Real-time telemetry platforms | 10,000+ concurrent devices held at P95 under 200 ms | ANZ | measured in production |
| Cloud-native migration — .NET, Azure | >30% delivery efficiency · >15% infrastructure cost reduction | ANZ | measured in production |
| Programme and portfolio management | $5M+ portfolio · 5+ squads · up to 40 practitioners onshore and offshore | ANZ | measured in production |
| Multi-agent system design — Python, FastAPI, Postgres, Redis | 20 agent engines · 22 routers · 4,272 backend tests · live on a VPS | aether-job-career-agent | measured in production |
| LLM evaluation and guardrails — Langfuse, Phoenix | −38% error-budget breaches; entailment guard reverts unsupported claimsthe −38% was measured against a simulated error budget, not live traffic | Independent · aether-job-career-agent | measured outside production |
| Next.js and TypeScript at production scale | 39 page routes · 2,326 unit cases · 26 Playwright specs | aether-job-career-agent | measured in production |
| Node.js / Express services | PEM key-distribution service · full Mocha/Chai coverage | public-key-server | measured in production |
| Containerised delivery — Docker, systemd, self-hosted CI | build → gate → deploy → smoke test → automatic rollback on failure | aether-job-career-agent · abentertainment | measured in production |
| Multi-service orchestration — Docker Compose | gateway, AI service and Redis composed across separate stacksCompose, not Kubernetes — there are no cluster manifests in that repository | containerised-birth-time-rectifier | measured outside production |
| WebGL and GLSL — three.js, React Three Fiber | bespoke shaders and scenes, one context per section, no context loss | this site · abentertainment | measured in production |
| Data visualisation | customer-journey timeline in React/TypeScript; sprint-velocity dashboard in Python | relationship-timeline-feature · EFDDH-Jira-Analytics-Dashboard | measured in production |
| Certified Scrum Master (CSM) | Scrum Alliance credential, held | Scrum Alliance | measured in production |
| Master of Computer Science (Honours) | conferred | Monash University | measured in production |
| Bachelor of Engineering, Computer Science | conferred | University of Melbourne | measured in production |
| AWS and GCP certification | studying; no certificate issuedlisted because the CV lists it — saying so is the point of the card | — | in progress, not yet held |
Every calibration above, on one page — exactly as it reads on the CV.
Download CVWhat is keeping me busy
Thirty-eight public repositories exist. These six are the ones worth your time, each with what it does not do printed beside what it does.
A job-application platform whose guard reverts any claim your own résumé cannot support.
Schematic · not live activity. Claim text passes through a résumé support guard; unsupported output returns to a revert path.
Source: plate description and limit for aether-job-career-agent.
Limit: the public CI workflow is red on main; production deploys through a separate gated pipeline.
Highlighted stage: A candidate claim enters the support check.
LimitsThe public CI workflow is red on main; production deploys through a separate gated pipeline.
A live event company’s site and admin portal, containerised and deployed from a push.
Schematic · not live activity. A push rebuilds the containerised site/admin path and returns to the published site.
Source: plate description and limit for abentertainment.
Limit: content persists as flat JSON files, not a database — sized for one editor, not many.
Highlighted stage: A repository push begins the rebuild path.
LimitsContent persists as flat JSON files, not a database — sized for one editor, not many.
An agent loop whose only exit is a signed verifier saying the work actually passed.
Schematic · not live activity. Work loops until the verifier signs an exit condition.
Source: plate description and limit for ralph-loop-infinite.
Limit: a harness for his own machine, not a product; it assumes a trusted local environment.
Highlighted stage: The local harness performs another work pass.
LimitsA harness for his own machine, not a product; it assumes a trusted local environment.
Rebuilds a raw prompt into a specification, with provider failover when a model declines.
Schematic · not live activity. A raw prompt is reconstructed into a specification, with provider failover when a model declines.
Source: plate description and limit for prompt-reconstruction-engine.
Limit: reconstruction quality is judged by the operator; there is no automated benchmark yet.
Highlighted stage: The operator’s original prompt starts the reconstruction.
LimitsReconstruction quality is judged by the operator; there is no automated benchmark yet.
A Vedic astrology API with an ephemeris accuracy gate that fails the build on drift.
Schematic · not live activity. The API computes positions and applies an ephemeris accuracy gate; this is not an astrological reading.
Source: plate description and limit for jyotish-shastra.
Limit: an engine, not an interpretation: it computes positions and declines to tell fortunes.
Highlighted stage: Ephemeris data is the accuracy reference.
LimitsAn engine, not an interpretation: it computes positions and declines to tell fortunes.
The page you are reading: static export, one WebGL context per section, no analytics.
Schematic · not live activity. A static page rail holds six sections and quoted figures; nothing is computed live for the reader.
Source: plate description and limit for forgotten-mistory.
Limit: every figure on it is quoted from a CV or a repository — none is computed live.
Highlighted stage: The page is shipped as a static export.
LimitsEvery figure on it is quoted from a CV or a repository — none is computed live.
These six are shipped work. The inbox that answers a role enquiry answers a project brief.
Email a 20-minute-call agenda38 public repositories · metrics harvested 2026-09-03 from the GitHub API, not live
Always willing to listen
I have been wrong often enough to want to hear it early. Tell me what you think — I’ll listen properly.
Resonance neutral — focus or hover a real contact route.
Coffee · Melbourne CBD · I’ll come to you