Melbourne, Australia

Lachlan
Stedman.

BI consultant at Acquum Consulting and builder of production AI systems. I deliver Infor EPM reporting and reconciliation for consulting clients, and I build agent systems that run real operations: an autonomous marketing pipeline, an MCP bridge that explains ledger variances, automation for clinics.

Commerce and computer science, La Trobe University. One half builds the system, the other half reads the numbers it produces.

Email meSee the experience
Why work with me

The expensive habit in AI systems: calling a model where a script would do.

Every unnecessary model call costs money, adds latency and creates a fresh chance to be wrong. I build the other way round. Deterministic code does the sourcing, scoring, state and safety gates, and the model is reserved for the judgement only it can make.

Every claim a system produces comes from measured data and is re-verified before it goes out. Every experiment is scored on outcomes. If those outcomes show an approach is not working, you hear it from me early, with the evidence, rather than after the budget is spent. The discipline that keeps costs flat is the same discipline that keeps results honest.

What you get is automation a finance team can sign off on: read-only by default, every call logged, cost per outcome measured rather than guessed.

Infor EPM & OLAPSQLPythonLLM agent systemsMCPToken & cost optimisationReconciliation & data qualityTypeScript & ReactServerless & Workers
Strengths

What clients get.

Reporting that ties out

Infor EPM, OLAP and SQL delivered end to end, with reconciliation checks and data-quality validation built into the staging layer rather than bolted on at the end.

Agents you can govern

Read-only by default, confirmation before any write, every call logged with secrets redacted. Automation a finance team can sign off on.

Cost discipline

Token and cost optimisation as a design habit: code before models, context engineering, caching, and cost per outcome measured rather than guessed.

Claims that survive checking

Nothing leaves a system unverified: send-time re-checks, circuit breakers, rolling-window health metrics, and honest reporting when the numbers disappoint.

Fluent in finance and engineering

Requirements from a finance team become working reports and controls, iterated with the stakeholders in the room on the real data issues rather than the tidy version.

Shipped end to end

From data model to serverless deployment, then operated in production: migrations, refunds, access control, backups, runbooks. The unglamorous parts included.

Relevant experience

Consulting delivery, Acquum Consulting.

Joined as a data analyst in October 2025 and moved into the consultant role within four months. Client names are withheld; the work is described by what was delivered.

Associate BI Consultant

February 2026 to current

Data Analyst

October 2025 to February 2026

Month-end reporting that reconciles

OLAP models, dashboards and workflow configuration for client finance teams on Infor EPM, with reconciliation checks and data-quality validation built into the SQL staging layer.

Agentic reconciliation, demonstrated to a client

ReconAgents, a hackathon build from August 2026: an in-tenant process flags ledger variances and Claude agents explain each one down to the missing journal lines. P&L drill-down demonstrated live.

Scheduled loads clients can rely on

Integrations and scheduled data loads supported and troubleshot, reliability improved, and repeatable run procedures documented so no client depends on one person's memory.

Expense and profitability reporting

OLAP models and dashboards validated against source data, with SQL staging tables and queries preparing the datasets behind them.

Requirements to working solution

Business requirements translated into technical designs and iterated with client stakeholders until the data issues were resolved and the reporting was actually used.

Environment support and training

Database backups and training on deliverables in the d/EPM environment, plus documentation of reporting logic, data definitions and run steps for handover.

Selected work

What I have built.

Hackathon build, August 2026

Agentic reconciliation for Infor EPM

Claude · MCP · Python · Infor EPM · Data Fabric · SunSystems

Month-end reconciliation means someone chasing why a ledger account is out. ReconAgents does the chasing. An in-tenant Infor EPM process flags variances at year, period and account grain, then Claude agents query EPM, Data Fabric and SunSystems through a Python MCP bridge and explain each variance down to the missing journal lines, in plain language.

  • Demonstrated live to a client
  • 8-check tie-out suite: a red row blocks sign-off
  • Read-only by default, OAuth2 through one gateway, every call logged
Operated live, August 2026

Agentic marketing pipeline

Python · SQLite · LLM agents · Cloudflare Workers & KV · Instantly API

A fully end-to-end agentic pipeline that sources leads from live APIs, scores them on measured site performance and writes personalised emails. When replies arrive the agent strips quoted text, classifies the response and surfaces the warm ones. A daily LLM project manager reports to Discord and runs A/B experiments on real funnel data.

Its own funnel data eventually showed the offer, not the system, was the constraint. I shut it down on the numbers, which is the point of having them.

  • 636 businesses sourced and scored
  • 143 personalised emails sent with no manual touches
  • Bounce rate healed from 9.1% to 0% by a self-correcting circuit breaker
Production system

Clinical practice automation

Playwright · Claude Haiku · Cloudflare Workers

Clinics lose staff hours to repetitive portal work. This multi-tenant system does it for them: Playwright browser automation at scale, driven by Claude Haiku, behind a Cloudflare Workers proxy that isolates each clinic's API key and meters its spend.

  • Multi-tenant with per-client key isolation
  • Per-client spend tracking
  • Small, cheap model chosen deliberately for the task
Knowledge tool

Self-learning documentation assistant

RAG · vector search · LLM APIs

Consultants spend hours hunting through Infor ERP documentation. This retrieval-augmented assistant answers from the knowledge base with vector search and context management, and improves its coverage as it is used.

  • Retrieval-augmented generation over Infor ERP documentation
  • Vector search with context management
  • Built for the ecosystem I work in daily
Quantitative system

Market-neutral funding-rate bot

Python · exchange APIs · risk modelling

Harvests perpetual-swap funding payments with zero directional exposure: long spot, short perp. No prediction anywhere in the codebase. It measures the yield that exists now, nets out fees and slippage, and uses entry and exit hysteresis so churn never eats the edge.

  • Paper-trading gated before any live capital
  • Fees and slippage netted before any position
  • Hysteresis on entry and exit
Product with real users

Payments-enabled web product

Firebase · Cloud Functions · Stripe · Firestore

A full product with real users and real money: Firebase Cloud Functions backend, Stripe checkout, billing portal and webhooks, Discord account linking and admin tooling. Built, shipped and operated, including migrations, refunds and access control.

  • Stripe checkout, billing portal and webhooks
  • Account linking and admin tooling
  • Operated: migrations, refunds, access control
01 / 06
Education & recognition
Bachelor of Commerce and Computer Science

La Trobe University · 2022 to 2025

Computer science capstone: a cross-platform contractor marketplace in Flutter with a Firebase backend, delivered by an Agile team through to app-store deployment. Finance capstone: an Australian REIT portfolio built, managed and defended to assessors.

Team Award, La Trobe Engineering & IT Showcase 2025

Awarded for the Scrum-developed Flutter marketplace app.

Ongoing
Anthropic coursework

March 2026 to current

RAG project development

January 2026 to current

Azure fundamentals and cloud concepts

November 2023 to November 2024

Working on something in AI or data?I answer my own email.

Email meOr try the live audit tool