Skip to content
Vikram Deshpande

ATO Payday Super · 200+ SIT scenarios

1 cell = 1 scenario · fill = effort
One cell for each of 200 scenarios. The fill is the effort for one: about 3 hours, falling to about 15 minutes.

≈92% less evidence effort per scenario

Architected COBOL/mainframe test-evidence automation for 200+ SIT/E2E scenarios across all eight Payday Super squads: ~3 hours to ~15 minutes per scenario. The REXX, SMF, SDSF, PCOMM, PowerShell and VBA toolchain needed no new approvals. Convened a cross-discipline war room that produced a binding automation recommendation in under three hours.

ANZ · up to 40 practitioners

1 mark = 1 practitioner
One mark for each practitioner, up to 40, across 5+ squads.

$5M+ portfolio across 5+ squads

Directed a program portfolio valued at over $5M at ANZ: 5+ cross-functional squads, up to 40 practitioners, onshore and offshore, delivering on-time, high-quality releases. Defined the technical vision and owned the product backlog for major platform modernisations, at 100% compliance with enterprise architectural standards and risk models.

P95 under 200 ms · ANZ

1 point = 1 device
One point for each of 10,000 concurrent devices, in blocks of 100. The line crosses them in 200 ms, the ceiling on P95 latency.

10,000+ concurrent devices

Led the technical delivery of AI solutions at ANZ, including real-time WebSocket telemetry servers for 10,000+ concurrent devices. The telemetry cut P95 latency to under 200 ms, improving user response time.

Experience

Eight roles

ATO2026–present8 months

Scrum Master / PM

≈92% less evidence effort across 200+ SIT scenarios

1 cell = 1 scenario · fill = effort

MYOB · Melbourne, VIC · May 2010 – August 2011

1 year 3 months

Reporting time cut 90% on financial data sets

1 cell = 1 month

Developer Support / Software Testing / Analyst

Developer support, software testing and data-processing optimisation.

  • Optimised data-processing workflows through ePAL implementation, improving efficiency by 30% and cutting reporting time by 90% for financial data sets.

InfoCentric · Melbourne, VIC · August 2011 – November 2014

3 years 3 months

Engagement up 20%; regulatory reporting automated to 100% accuracy

1 cell = 1 month

Senior Business Analyst

Analytics and business intelligence delivery with automated regulatory reporting.

  • Delivered analytics and business intelligence projects that lifted client engagement by 20%, and automated regulatory reporting workflows to 100% accuracy.

Telstra · Melbourne, VIC · November 2014 – October 2015

11 months

Delivery efficiency up 20%, operational clarity up 15%

1 cell = 1 month

Business Analyst / Project Coordinator

Customer-journey scorecards and JIRA requirements for core delivery teams.

  • Developed customer-journey scorecards and streamlined JIRA requirements for core delivery teams, improving delivery efficiency by 20% and operational clarity by 15%.

Microsoft · Sydney, NSW · October 2015 – October 2016

1 year

Raised reliability 15% and cut incident MTTR 10%

1 cell = 1 month

Lead Business Analyst

Azure ML telemetry gap analysis and DevOps alignment with enterprise architecture.

  • Executed a gap analysis for Azure ML telemetry that delivered 10 key insights, raised system reliability by 15% and cut incident MTTR by 10%.
  • Aligned DevOps strategies with overall IT objectives, reaching 95% alignment with enterprise architecture standards.

National Australia Bank (NAB) · Melbourne, VIC · November 2016 – September 2017

10 months

100% regulatory adherence across major data initiatives

1 cell = 1 month

Senior Project Manager and Business Analyst

Delivery stream for a critical risk and compliance program.

  • Managed the delivery stream of a critical risk and compliance program, ensuring 100% regulatory adherence for major data initiatives.
  • Conducted deep-dive business analysis, translating complex regulatory requirements into actionable technical specifications for development teams.

ANZ · Melbourne, VIC · September 2017 – June 2025

7 years 9 months

10,000+ concurrent devices at P95 under 200 ms

1 block = 100 devices

Senior Delivery Lead / Technical Product Owner

AI/ML Strategy & Solutions Architect, 2017 – 2022

Delivery lead and AI/ML architect for core banking modernisation and real-time telemetry.

  • Led the technical delivery of AI solutions from 2022 to 2025, including real-time WebSocket telemetry servers for 10,000+ concurrent devices that cut P95 latency to under 200 ms and improved user response time.
  • Moved core banking platforms from monolith to cloud-native (.NET/Azure) between 2017 and 2022: end-to-end delivery time down by more than 30%, infrastructure cost down by more than 15%.
  • Directed a program portfolio valued at over $5M across 5+ cross-functional squads and up to 40 practitioners, onshore and offshore, to on-time, high-quality releases.
  • Owned the technical vision and product backlog for major platform modernisations at 100% compliance with enterprise architectural standards and risk models.
  • Facilitated workshops for 40+ GMs and executives to align on strategy, lifting decision-making efficiency and project clarity by more than 55%.

Independent AI Consulting and Upskilling · Melbourne, VIC · June 2025 – February 2026

8 months

−38% error-budget breaches (in simulation)

1 cell = 1% of breaches

AI Solutions Consultant, Applied LLM Engineering and Delivery Tooling

Applied LLM engineering: evaluation tooling, delivery analytics and autonomous agents.

  • Architected a Next.js and Supabase analytics dashboard for JIRA that exposes sprint velocity metrics and generates LLM-powered retrospective insights and sprint plans.
  • Developed a React/TypeScript visualisation tool with D3 event arcs that showcases customer-journey interactions.
  • Stood up a Langfuse and Phoenix evaluation stack that scores hallucination, latency and cost across high-volume daily LLM generations: −38% error-budget breaches (in simulation).
  • Built a production-grade Node.js/Express service that distributes PEM keys for fleet API signing, with 100% test coverage (Mocha/Chai).
  • Built AI Resume Tailor, an NLP system that uses web scraping and prompt engineering to match CVs to job descriptions.
  • Created AI Gmail Manager, an autonomous TypeScript agent for inbox triage, sentiment analysis and draft generation.
  • Developed Jyotish Shastra, a Vedic astrology API with an ephemeris accuracy gate, and BTR-Demo, which explores Vedic algorithms with AI/ML.
  • Started the @vicd0ct YouTube channel: technical deep-dives on live coding, algorithm archaeology and telemetry breakdowns.

Australian Taxation Office (ATO) · Melbourne, VIC · March 2026 – present

8 months

≈92% less evidence effort across 200+ SIT scenarios

1 cell = 1 scenario · fill = effort

Scrum Master / Project Manager, Payday Super program (NTP and Distribution UI)

Leads one of eight Payday Super squads; architected the program’s test automation.

  • Leads the Kookaburras squad, one of eight squads on the Payday Super program: sprint cadence, PI planning (PI 47–48), capacity management and executive status reporting.
  • Architected COBOL/mainframe test-evidence automation for 200+ SIT/E2E scenarios across all eight squads: ~3 hours to ~15 minutes per scenario, a ≈92% reduction, on a zero-new-approvals toolchain (REXX, SMF, SDSF, PCOMM, PowerShell, VBA).
  • Turned an infeasible SIT window (75+ hours of manual evidence per team against 64 available) into an achievable plan: a six-day tiered harness build, a formal go/no-go gate and a four-level contingency ladder.
  • Authored the executive change request re-baselining Payday Super test capacity from 30 to up to 90 person-days: options analysis, costed recommendation and risk assessment.
  • The request carried full Azure DevOps traceability across 40 scenarios and 11 data tables (AC6–AC19), with STT regression scope.
  • Convened a cross-discipline war room that produced a binding automation recommendation in under three hours.
  • Unblocked stalled NTP function testing through L2 environment escalation and targeted SME enablement, and steered Distribution UI delivery to 95%+ completion.

Work

Selected work

Live

  1. A candidate claim enters the support check.

  2. An entailment check admits the claims the résumé supports.

  3. Unsupported output returns to the revert path instead of publishing.

Aether

A multi-agent job-application platform: 20 agent engines behind 22 routers, with a Next.js front end of 39 page routes. An entailment guard reverts any claim the résumé cannot support.

agent engines
20
routers
22
page routes
39
backend tests
4,272
unit cases
2,326
Playwright specs
26

Stack

  • Python
  • TypeScript
  • JavaScript
  • HTML
  • FastAPI
  • Postgres
  • Redis
  • Next.js
  • Docker
  • systemd
  • Self-hosted CI

Engineering decisions

  • An entailment guard sits between generation and output, so a claim the résumé cannot support reverts.
  • An unmeasured signal reads “not measured”, never zero.
  • Releases run build, gate, deploy and smoke test, with automatic rollback on failure.
  • Production deploys through a separate gated pipeline.

First commit 1 July 2026, 1,666 commits as of 10 November 2026

Live

  1. A repository push starts the rebuild.

  2. The site and admin portal build as containers.

  3. The public event-company site is the result.

AB Entertainment

The site and admin portal of a live event company, packaged as containers and rebuilt from a repository push. The repository carries bespoke WebGL shaders and scenes.

Stack

  • HTML
  • TypeScript
  • JavaScript
  • Shell
  • WebGL
  • GLSL
  • Docker
  • systemd
  • Self-hosted CI

Engineering decisions

  • Content persists as flat JSON files, sized for a single editor.
  • Releases run build, gate, deploy and smoke test, with automatic rollback on failure.

First commit 28 March 2026, 499 commits as of 10 November 2026

Verifier-gated

  1. The local harness performs a work pass.

  2. A verifier checks whether the pass passed.

  3. A signed verifier result leaves the loop.

Ralph Loop

An agent loop that repeats a work pass until a verifier confirms that the work passed. A signed verifier result is the sole exit condition.

Stack

  • Shell
  • Python
  • HTML

Engineering decisions

  • A signed verifier result is the sole exit from the loop.
  • The harness runs in a trusted local environment.

First commit 19 May 2026, 31 commits as of 10 November 2026

Provider failover

  1. The operator’s original prompt starts the reconstruction.

  2. The engine reshapes the prompt into a specification.

  3. A model decline hands the request to failover.

Prompt Reconstruction

Rebuilds an operator’s raw prompt into a specification. When a model declines, provider failover takes over.

Stack

  • HTML
  • JavaScript

Engineering decisions

  • Provider failover keeps reconstruction running when a model declines.
  • The operator judges reconstruction quality.

First commit 16 June 2026, 34 commits as of 10 November 2026

Accuracy-gated

  1. Ephemeris data is the accuracy reference.

  2. The build fails on drift beyond the gate.

  3. The API returns computed positions.

Jyotish Shastra

A Vedic astrology API that computes positions from ephemeris data. An accuracy gate fails the build on drift.

Stack

  • JavaScript
  • CSS
  • Shell
  • TypeScript

Engineering decisions

  • An ephemeris accuracy gate fails the build on drift.
  • Positions are computed; interpretation is out of scope.

First commit 24 June 2025, 214 commits as of 10 November 2026

Live

  1. Content lives in typed TypeScript files.

  2. Next.js exports the pages as static files for Firebase Hosting.

  3. One script checks, builds, tests, pushes, deploys and verifies the live build.

Forgotten Mistory

A portfolio built as a static export on Firebase Hosting. Typed content files feed every page, and one release script checks, builds, tests and deploys.

Stack

  • TypeScript
  • JavaScript
  • CSS
  • Python
  • Next.js 15
  • React 19
  • Tailwind CSS 4
  • Firebase Hosting

Engineering decisions

  • No analytics and no cookies; a page load writes nothing to storage and contacts one host.
  • A strict Content-Security-Policy is enforced by an automated test.
  • Content is typed, and the test suite checks its shape and wording before every release.

First commit 25 November 2025, 874 commits as of 10 November 2026

Capabilities

Skills, linked to where they were applied

Delivery leadershipAI systemsCloud and platformProduct engineeringQualificationsAgile delivery at scaleProgram and portfolio managementMainframe test automationCertified Scrum MasterMulti-agent systemsLLM evaluation and guardrailsReal-time telemetryCloud-native migrationContainerised deliveryService orchestrationNode.js servicesAWS and GCP (in progress)Next.js and TypeScriptWebGL and GLSLData visualisationMaster of Computer Science (Honours)Bachelor of Engineering, Computer ScienceATO Payday Super (program)Scrum Alliance (qualification)ANZ (program)Independent consulting (program)aether-job-career-agent (repository)containerised-birth-time-rectifier (repository)public-key-server (repository)abentertainment (repository)relationship-timeline-feature (repository)EFDDH-Jira-Analytics-Dashboard (repository)Monash University (qualification)University of Melbourne (qualification)

Delivery leadership

  • ATO Payday Super, program
  • ANZ, program
  • Scrum Alliance, qualification

AI systems

  • aether-job-career-agent, repository
  • Independent consulting, program
  • ANZ, program

Cloud and platform

  • ANZ, program
  • aether-job-career-agent, repository
  • abentertainment, repository
  • containerised-birth-time-rectifier, repository
  • public-key-server, repository

Product engineering

  • aether-job-career-agent, repository
  • abentertainment, repository
  • relationship-timeline-feature, repository
  • EFDDH-Jira-Analytics-Dashboard, repository

Qualifications

  • Monash University, qualification
  • University of Melbourne, qualification
  • Program
  • Repository
  • Qualification
  • In progress

Delivery leadership

Delivery leadershipAgile delivery at scaleATO Payday Super

Agile delivery at scale

At the ATO, leads the Kookaburras squad, one of eight Payday Super squads, and owns its PI planning (PI 47–48). Authored the change request that re-baselined test capacity from 30 to up to 90 person-days.

Delivery leadership

Delivery leadershipProgram and portfolio managementANZ

Program and portfolio management

At ANZ, a $5M+ portfolio across 5+ squads and up to 40 practitioners, onshore and offshore.

Delivery leadership

Delivery leadershipMainframe test automationATO Payday Super

Mainframe test automation

At the ATO, REXX, SMF, SDSF and PCOMM automation took evidence effort from ~3 hours to ~15 minutes per scenario across 200+ SIT/E2E scenarios.

Delivery leadership

Delivery leadershipCertified Scrum MasterScrum Alliance

Certified Scrum Master

Certified Scrum Master (CSM), Scrum Alliance.

Applied at

  • Scrum AllianceQualification

AI systems

AI systemsMulti-agent systemsaether-job-career-agent

Multi-agent systems

aether-job-career-agent: 20 agent engines and 22 routers on Python, FastAPI, Postgres and Redis, with 4,272 backend tests, live on a VPS.

AI systems

AI systemsLLM evaluation and guardrailsIndependent consultingaether-job-career-agent

LLM evaluation and guardrails

A Langfuse and Phoenix evaluation stack cut error-budget breaches by 38% in simulation; in Aether, an entailment guard reverts unsupported claims.

AI systems

AI systemsReal-time telemetryANZ

Real-time telemetry

At ANZ, WebSocket telemetry cut P95 latency to under 200 ms across 10,000+ concurrent devices.

Cloud and platform

Cloud and platformCloud-native migrationANZ

Cloud-native migration

At ANZ, core banking platforms moved from monolith to .NET/Azure: end-to-end delivery time down by more than 30%, infrastructure cost down by more than 15%.

Cloud and platform

Cloud and platformContainerised deliveryaether-job-career-agentabentertainment

Containerised delivery

aether-job-career-agent and abentertainment: Docker, systemd and self-hosted CI run build, gate, deploy and smoke test, with automatic rollback on failure.

Cloud and platform

Cloud and platformService orchestrationcontainerised-birth-time-rectifier

Service orchestration

containerised-birth-time-rectifier: a gateway, an AI service and Redis composed with Docker Compose across separate stacks (non-production).

Applied at

  • containerised-birth-time-rectifierRepository

Cloud and platform

Cloud and platformNode.js servicespublic-key-server

Node.js services

public-key-server: a Node.js/Express service that distributes PEM keys for fleet API signing, with 100% test coverage (Mocha/Chai).

Applied at

  • public-key-serverRepository

Cloud and platform

Cloud and platformAWS and GCP (in progress)

AWS and GCP (in progress)

Cloud and data certifications with AWS and GCP are in progress.

Product engineering

Product engineeringNext.js and TypeScriptaether-job-career-agent

Next.js and TypeScript

aether-job-career-agent: 39 page routes, 2,326 unit cases and 26 Playwright specs.

Product engineering

Product engineeringWebGL and GLSLabentertainment

WebGL and GLSL

abentertainment: bespoke shaders and scenes.

Product engineering

Product engineeringData visualisationrelationship-timeline-featureEFDDH-Jira-Analytics-Dashboard

Data visualisation

relationship-timeline-feature: a customer-journey timeline in React/TypeScript with D3 event arcs. EFDDH-Jira-Analytics-Dashboard: a sprint-velocity dashboard in Python.

Applied at

  • relationship-timeline-featureRepository
  • EFDDH-Jira-Analytics-DashboardRepository

Qualifications

QualificationsMaster of Computer Science (Honours)Monash University

Master of Computer Science (Honours)

Monash University.

Applied at

  • Monash UniversityQualification

Qualifications

QualificationsBachelor of Engineering, Computer ScienceUniversity of Melbourne

Bachelor of Engineering, Computer Science

University of Melbourne.

Applied at

  • University of MelbourneQualification

About

How I work

Portrait of Vikram Deshpande

At the ATO I lead the Kookaburras squad, one of eight Payday Super squads. At ANZ I led AI/ML delivery and core banking modernisation; on my own, I build multi-agent systems. Next: programs where AI delivery and AI assurance are the hard part.

  • Architect the delivery

    Delivery and architecture are one job.

    At the ATO, the eight-squad test-evidence problem was a delivery problem solved with a toolchain.

    ATO · 200+ SIT scenarios
  • Practise agile

    Cadence and PI planning when the plan holds; a cross-discipline war room in under three hours when it does not.

    ATO · Payday Super
  • Own AI assurance

    Accountable for whether model output can be trusted, as well as whether it shipped.

    Langfuse and Phoenix · −38% (in simulation)
  • State the number

    A status report can carry bad news; every program I have rescued turned on someone stating the number early.

    75+ hours of evidence against 64 available: escalated, then re-baselined.

    ATO · SIT window
  • Measure, then claim

    In Aether, market figures cite their source, and an unmeasured signal reads “not measured”, never zero.

    aether-job-career-agent

Education and certifications

  • Master of Computer Science (Honours)

    Monash University

  • Bachelor of Engineering, Computer Science

    University of Melbourne

  • Certified Scrum Master (CSM)

    Scrum Alliance

  • AWS and GCP

    Certification in progress

Contact

Start with a 20-minute call

Open to Scrum Master / Project Manager roles in Melbourne and selected AI delivery engagements.