RMC.
Home  /  Companies  /  Databricks
Industry View →
CSQ227R47 · Verified live · 04 May 2026

AI Engineer · FDE

Forward Deployed Engineer, AI Forward Deployed Engineering team. Melbourne, Australia. Open to remote locations. Customer-facing: build and productionize first-of-its-kind GenAI applications alongside Databricks customers.

Company Databricks Inc. Stage $134B (Series L · Dec 2025) · IPO-ready Run-rate US$5.4B · +65% YoY Customers 20,000+ · 70% of Fortune 500 Reports to AI FDE leadership · ANZ region Apply via Greenhouse — employer-direct

01Mandate vs Approach

five role pillars · paired
Databricks asks
Develop cutting-edge GenAI solutions incorporating the latest from Mosaic AI Research — RAG, multi-agent systems, Text2SQL, fine-tuning — using HuggingFace · LangChain · DSPy.
What I bring
27 years building centralised business systems from scratch. The DOT (Demand & Operations Toolkit) architecture at Saputo Dairy Australia was a 7-layer design on SAP ECC modelling the full Australian operation across 970+ materials, 231 Power BI visuals, 155 DAX measures. Solo build over nine years on personal equipment; layers 5–7 were built with Anthropic's Claude as engineering partner from 2024 (RAG patterns over the operational data, prompt chains for exception handling, agentic workflows for cross-layer reconciliation). The system was not adopted at Saputo; the AI-assisted personal-equipment build was the integrity issue I voluntarily disclosed to corporate security on exit. The toolchain is different; the build pattern is identical.
Databricks asks
Own production rollouts of consumer and internal-facing GenAI applications end-to-end.
What I bring
End-to-end ownership is what I do. Pacific Brands Workwear Group, 2011–2013 — group-level role, three management tiers above where most builders sit. Built a centralised VBA engine harvesting six ERPs (Pronto, MYOB, JDE, two SAP variants, NetSuite-equivalent), auto-generating supplier orders, shipping branded dashboards per division, and producing S&OP decks for CEOs with mandatory sign-off governance. Multi-site team across VIC/NSW, extended to NZ. End-to-end ownership from requirement to production rollout — across an otherwise fragmented multi-division business.
Databricks asks
Trusted technical advisor to customers across a variety of domains.
What I bring
Three "impossible" centralised systems built across three different industries — apparel manufacturing, defence apparel (military BOM complexity, expensive defence-grade inventory), dairy operations. The pattern transfers because I learn the customer's domain before I build. The DOT system at Saputo wasn't a Power BI dashboard exercise; it was 7 layers of integrated planning logic that 600+ users relied on daily because it actually solved their problem. That's what trusted technical advisor means.
Databricks asks
Present at conferences (Data + AI Summit), recognised as a thought leader internally and externally.
What I bring
The site you're reading is my external proof. robertmclark.dev is a Cloudflare Workers application built with Claude as engineering partner — themed company dashboards, industry briefings, and a personal pitch system. It demonstrates the methodology before any conversation starts. Writing and presenting on Forward Deployed Engineering — what it actually means versus what it sounds like — is something I'd do gladly. Plain English on AI build patterns is in short supply.
Databricks asks
Collaborate cross-functionally with product and engineering teams to influence priorities and shape the product roadmap.
What I bring
Pacific Brands had me sitting between six different ERP teams, four divisional GMs, three CEOs across Workwear / Underwear / Footwear, and a CFO who wanted one number, one truth. Saputo had me bridging operations, finance, supply, and IT to build the DOT system. Cross-functional is the only mode I work in. The deliverable I'm best known for is the artefact that makes a cross-functional team agree on what to do next.

02Why Robert M. Clark fits

four anchor angles
ANGLE 01

Forward Deployed Engineer is what I've been doing for 27 years — without the title

Embedded with the customer, owning the build end-to-end, presenting back to senior stakeholders, iterating in production. The label is new to my CV; the work isn't. Pacific Brands was FDE before FDE was a recognised role. Saputo's DOT system was FDE inside the customer organisation.

ANGLE 02

Plain-English bridge between business problem and AI solution

The Databricks Australian customer base — Coles, NAB, Macquarie, Telstra, Westpac NZ, Atlassian — is dominated by enterprises whose data and ML maturity is uneven. They need someone who speaks both supply chain and Python, both finance and prompt engineering. I'm closer to that bridge than most candidates with stronger ML CVs.

ANGLE 03

Two years of Claude-as-engineering-partner work, on enterprise-scale architecture

Layers 5–7 of the DOT architecture and the KIM pegging engine were built with Claude as engineering partner from 2024 onward. RAG patterns, prompt chains, agentic exception handling, multi-step reconciliation. Not toy projects — enterprise-scale build patterns at the limits of what one person + AI can architect. The methodology is what Databricks customers are asking their FDEs to teach them.

ANGLE 04

My specialisation is the slot the ensemble is missing

The role description explicitly says you build the AI FDE team as an ensemble of unique specialisations. Mine: enterprise-customer-embedded planning systems delivery, three industries deep, with two years of Claude-as-build-partner methodology layered on. Most strong AI engineers don't have the customer-side enterprise spine. Most strong customer-side enterprise builders don't have the GenAI methodology yet. The overlap is small — and that's where I sit.

03Skills match

requirement → evidence → fit
Databricks requirement
Evidence in my track record
Match
RAG, multi-agent systems, Text2SQL, fine-tuning
DOT layers 5–7 use RAG patterns over operational metadata. KIM pegging engine uses agentic exception handling. Text2SQL pattern for natural-language ops queries. Two years on these patterns with Claude as partner.
Direct
HuggingFace · LangChain · DSPy tooling
Working knowledge of the patterns rather than years on each library. Comfortable picking up specific framework idioms quickly — that's what 27 years across Pronto / MYOB / JDE / Dynamics AX / SAP ECC / Power Platform looks like.
Stretch
Production-grade GenAI deployments with eval & optimisation
Saputo DOT runs daily for 600+ users. KIM pegging runs on millions of order lines. Production discipline (eval, monitoring, rollback, change control) is non-negotiable in supply chain — same principles apply to GenAI systems.
Direct
Python data science stack: pandas, scikit-learn, PyTorch
Stack has been Microsoft-side: Power Query M, DAX, VBA, Power Automate, T-SQL. Python familiarity for data work, not deep ML training experience. Learning curve is real and acknowledged — but pattern recognition transfers.
Stretch
Production ML on AWS, Azure, or GCP
Cloud deployment via Cloudflare Workers (robertmclark.dev), Azure SharePoint integration on the DOT system, Microsoft 365 Power Platform stack throughout Saputo. Comfortable in cloud; less battle-tested specifically on AWS GenAI primitives.
Adjacent
Communicating technical concepts to non-technical and technical audiences
CEO sign-off governance at Pacific Brands. 600+ users interacting daily with the DOT system. The site itself is the proof artefact. Plain English is the discipline I optimise for.
Direct
Apache Spark / Databricks Intelligence Platform [preferred]
No Databricks production experience. Familiar with the architecture (Lakehouse, Delta Lake, MLflow, Unity Catalog, Mosaic AI Research) at a working level. First 30 days would be platform-fluency closing.
Stretch
Graduate degree in quantitative discipline OR equivalent practical experience
No formal graduate degree. Twenty-seven years of customer-embedded enterprise build across six ERP environments, three industries, and two centralised platforms (Pacific Brands group-level engine; Saputo DOT architecture with Claude as engineering partner) — the "equivalent practical experience" the role explicitly accepts.
Direct

04Build proof

three centralised systems · three industries
Crown jewel · Pacific Brands Workwear Group

Six-ERP centralised planning engine

2011–2013 · Group-level · VIC / NSW / NZ · CEO governance

Operating at Workwear Group level — three management tiers above where I sit today — built a single planning engine across the entire group. Harvested data from six divisional ERPs, auto-generated supplier orders, produced branded dashboards per division and S&OP decks for the three divisional CEOs with mandatory sign-off governance. Extended into NZ inside the original delivery window.

The relevant point for Databricks: this was Forward Deployed Engineering before the term existed. Embedded inside the customer organisation, owning the build, presenting back to executives, iterating in production.

6ERPs unified
3Divisions · CEO sign-off
3Regions VIC/NSW/NZ
1Builder
Recent build · Saputo Dairy Australia

DOT — 7-layer planning architecture

2016–2025 · Australian operation · SAP ECC · Power BI

Built the DOT (Demand & Operations Toolkit) reporting system over nine years, supporting the full Australian operation. Seven integrated layers covering demand sensing, master scheduling, supply planning, inventory deployment, financial reconciliation, and exception management. Layers 5–7 (the most recent) were built with Anthropic's Claude as a genuine engineering partner — not Microsoft Copilot autocomplete, full architectural collaboration.

Used by 600+ planners, schedulers, supply analysts and operations leaders every working day. The methodology is the methodology Databricks customers buy FDE engagements to learn.

970+Materials managed
231Power BI visuals
155DAX measures
600+Daily users

05Risks I'd raise myself

honest framing · before they ask
Python ML depth
My production stack has been Microsoft / VBA / DAX / SQL. Comfortable in Python; not a years-deep ML engineer on PyTorch / scikit-learn workloads. Honest framing: first 60 days would include closing the platform-specific gaps. The pattern recognition transfers; the syntax is the work.
Databricks platform fluency
No production Databricks experience. Familiar with the architecture and product line at a working level (Lakehouse, Delta Lake, MLflow, Unity Catalog, Agent Bricks, Genie, Lakebase). Honest framing: ramp inside the FDE onboarding programme. I close platform gaps fast — that's what 27 years across six ERPs trains.
Career direction
I deliberately stepped down from group-level management to IC builder. Not because I couldn't — because building is what I'm good at and what I want to spend the next decade doing. Honest framing: if Databricks needs an FDE who'll still want to build in three years, that's me.

06Industry pulse

data + AI platforms · what an FDE leader watches
DBX revenue run-rate $5.4B +65% YoY
DBX valuation (Series L) $134B Dec 2025
DBX AI-product run-rate $1.4B ~26% of total
Snowflake (NYSE:SNOW) $58B market cap · trailing
Customers — 70% of Fortune 500 20,000+ across regions
$1M+ ARR customers 650+ land & expand
IPO timeline 2H 2026 S-1 expected Q3
Australian regulatory load CPS 234 + SOCI · Privacy · CDR
Ready to talk about whether the FDE pattern, customer-side, fits Databricks ANZ. Plain English, sharp edges, with the evidence attached.
robertmclark.dev · /company/databricks.html · v17.0 Robert M. Clark · Warrnambool VIC · AEST · Remote since 2019