Back to Portfolio
Project · Data Platform Build · 12 months

Designing an end-to-end data intelligence platform for a mid-market trading company.

A one-year project to take a growing trading company from scattered operational systems to a unified data platform — Data Lake, Data Warehouse, BI, and predictive models, all built from scratch. This was the foundation engagement where I formalized the planning cadence and sprint discipline I still use today.

RoleProject Manager
Duration12 months
Data Sources3 systems
Stack LayersDL · DW · BI · ML
Project ManagementScrum + KanbanModern Data StackData LakeData WarehouseBI & Predictive ModelingCloud (AWS / GCP / Azure)
01 — Context

From scattered operational data to a single source of analytical truth.

The starting position

The client — a mid-market trading company — was running their business on three operational systems that didn't talk to each other. Reports were assembled manually, leadership decisions ran on stale numbers, and there was no way to look forward: no forecasting, no demand modeling, no predictive view of the business.

The brief was deliberately ambitious. Build the entire analytical stack — ingestion, lake, warehouse, BI, predictive models — inside a single 12-month engagement, while delivering visible value along the way so leadership could feel the program working before it finished.

The challenge

One year. Four phases. Continuous value. Pure waterfall would have given leadership nothing to look at for nine months. Pure agile without a plan would have lost the architectural coherence the platform needed.

The answer was a hybrid: a high-level annual plan locking the macro phases and milestones, combined with two-week sprint execution inside each phase. Strategy at the top. Iteration at the bottom. The plan you'll see below is exactly that — and it's the planning model I still anchor delivery work to today.

02 — Annual Lifecycle

Four quarters. Four phase shifts. One continuous build.

Each quarter has its own dominant phase, but the transitions are soft — planning continues into execution, monitoring begins before deployment ends. The boundaries below show emphasis, not isolation.

Infochart 01 · Annual Timeline
Four phases mapped across the 12-month engagement
Annual Project TimelineA timeline showing four quarters and four project phases across a 12-month engagement: Planning (Q1), MVP Execution (Q2), Deploy & Test (Q3), and Monitor & Report (Q4), each with key activities listed.Quarter 1Quarter 2Quarter 3Quarter 4StartClose01020304Planning• Business & technical reqs• Cloud provider selection• Architecture for DL/DW + BI• Access & governance policy• Data dictionary + modelingMVP Execution• Data Lake ingestion• BI MVP build• DW implementation + tests• Validation + BI feedback• Predictive model setupDeploy & Test• Predictive modeling tests• Final predictive deploy• End-to-end integration testsMonitor & Report• Solution monitoring• Training & enablement• Feature expansion• Final report & sign-off
03 — Quarterly Goals

One concrete milestone per quarter, visible from outside the team.

Internal velocity is not a metric the client can feel. Every quarter therefore had to end on something demonstrable — something a stakeholder could see, click, or query.

Q1 · Foundation

Data flowing into the lake

Diagnostic + system spec complete; cloud provider selected; first ingestion pipelines live with data landing in the raw layer.

Data in raw layer
Q2 · Insight

BI MVP, DW operational

Warehouse built and validated; BI MVP shipped to business users; predictive modeling foundations laid for Q3.

BI MVP delivered
Q3 · Predict

Predictive solution live

Modeling validated, integrated tests passed, predictive solution deployed into production with monitoring.

Predictive solution shipped
Q4 · Sustain

Operational + opportunity report

Solution in steady-state monitoring; team trained; expansion roadmap delivered; two new features contracted.

Opportunities report delivered
04 — Backlog Model

Three levels of detail. Each one earns its place.

Coarse-grained items are for the client and the contract. Fine-grained items are for the team and the sprint. The middle layer is where they meet — and where the conversation about scope actually happens.

Infochart 02 · Backlog Hierarchy
Deliverable → Activity → Task
The top two levels live in the project backlog. The bottom level is what enters a sprint.
Backlog Hierarchy DiagramA three-level hierarchy: Deliverable (contractual unit the client signs off) at the top, two Activities (bodies of work) in the middle, and four Tasks (atomic sprint-level units) at the bottom. The upper two levels belong to the Project Backlog; tasks live in the Sprint Backlog.Project BacklogDeliverables & activitiesrefined with the team —the contractual outline.Sprint BacklogTasks the team commitsto inside a two-weeksprint window.DeliverableOutcome the client receivesActivity 1Body of workActivity 2Body of workTask 1Task 2Task 3Task 4Definition at each levelDeliverable. What the client signs off — the contractual unit.Activity. A body of work inside a deliverable; how the team groups effort.Task. An atomic unit a team member can absorb inside a sprint.
05 — Q1 Deliverables

Three deliverables that set every downstream decision.

Q1 is the most consequential quarter of the project — not because it ships the most, but because every Q2–Q4 decision references the artifacts produced here. Get these wrong and the rest of the year compounds the error.

Initial Diagnostic Report
Business + technical requirements, cloud cost estimates across AWS / GCP / Azure, architecture validation criteria, data dictionary, and a first modeling proposal.
System Specification
Final DL/DW requirements, access & governance policies, ingestion + processing architecture, initial infrastructure provisioning (buckets, clusters, VPNs, access controls).
Data Lake Ingestion
Ingestion services for all three source systems, transformation routines, dimensional + fact models in the warehouse, and a stakeholder workshop to validate the approach.
Acceptance Criteria (per deliverable)
Diagnostic → PM sign-off · System Spec → client approval · DL Ingestion → service in operation, validated against the spec.
06 — Sprint 1 · Resource Allocation

Sprint one was built tight on purpose.

A first sprint is also a calibration exercise. I planned Sprint 1 with no slack to measure real team capacity — then absorbed the learning into Sprint 2 onwards. The numbers below are the planned envelope; the lesson was always going to be in the variance.

Infochart 03 · Sprint 1 Allocation
Hours per role across activities (2-week sprint)
Sprint 1 Resource AllocationHours per role across four activities in the first two-week sprint. Business requirements: Data Scientist 20h, Data Engineer 5h. Technical requirements: Data Scientist 20h, Data Engineer 5h. Cloud cost and provider selection: Data Engineer 10h. Project kick-off: PM 10h. Sprint 1 total: 70 hours, planned with no buffer.ActivityData Eng. (h)Data Sci. (h)PM (h)01Business requirementsPrep + stakeholder interviews52002Technical requirementsPrep + tech-team interviews52003Cloud cost & provider selectionPricing study + proposal compilation1004Project kick-off (PM)Template, comms strategy, schedule10Sprint 1 totalsplanned with no slack, by design20h40h10hSprint 1 = 70h total committed across three roles, no buffer.The plan was to measure delivered vs. committed — and feed the gap into Sprint 2's capacity envelope. This calibration step is now standard in every engagement I lead.
07 — Sprint 1 · Daily Plan

What the team actually does on each day.

A weekly view that shows the rhythm: prep early in the week, stakeholder time mid-week, group discussion and documentation late in the week. Two weeks of this gets the foundational interviews and cloud proposal done end-to-end.

Infochart 04 · Two-Week Sprint Calendar
Daily activity by role
Two-Week Sprint CalendarDaily activity schedule across two weeks for three roles: Data Scientist, Data Engineer, and PM. Monday through Wednesday: preparation, cloud cost research, and PM template work. Thursday: stakeholder interviews (business interviews in Week 1, technical interviews in Week 2) followed by a cross-team group discussion. Friday: documentation for all roles.WEEK 1WEEK 2Data Sci.Data Eng.PMData Sci.Data Eng.PMMonPrep biz interviewCloud cost researchTemplate adaptationPrep tech interviewCloud cost researchTemplate adaptationTuePrep biz interviewCloud cost researchComms strategyPrep tech interviewCloud cost researchComms strategyWedPrep biz interviewCloud proposal draftProject schedulePrep tech interviewCloud proposal draftProject scheduleThuBusiness interviewBusiness interviewBusiness interviewTech interviewTech interviewTech interviewGroup discussion · share findings across teamGroup discussion · share findings across teamFriDocumentation of results · all rolesDocumentation of results · all rolesLegendPrep · researchProposal draftingPM activitiesStakeholder timeTeam syncScrum ceremonies (daily standup, planning, review, retro) sit outside this view — they're owned by the Scrum Master, not budgeted into role hours.
08 — Kanban Workflow

From the project backlog to “done” — six columns, one flow.

The board mirrors the backlog hierarchy. Deliverables and activities sit on the left as a roadmap; refined tasks flow rightward through the sprint columns. The Review column is the quality control mechanism — every task gets a second pair of eyes before it can be marked complete.

Infochart 05 · Kanban Board
Six-column flow with quality gate
Kanban Workflow BoardA six-column kanban board divided into two zones. The Roadmap zone on the left contains a Project Backlog column (with deliverables such as Initial Diagnostic, System Specification, and DL Ingestion) and a Sprint Backlog column (with refined tasks ready for sprint). The Sprint Execution zone on the right contains four columns: To Do, In Progress, Review (amber quality-gate column), and Done. Tasks move leftward to rightward through the sprint columns; every task must pass the Review quality-control gate before it can be marked Done.RoadmapProject BacklogInitial DiagnosticSystem SpecificationDL Ingestion+ activities (refined)Sprint BacklogPrep biz interviewsBiz interviewsCloud cost researchCloud proposalKick-off comms planSprint ExecutionTo DoTask: pull cost dataDE · 2hTask: define interview QDS · 15hIn ProgressTask: draft cloud specDE · 4h · in-flightTask: schedule biz mtgDS · 5h · in-flightReview (QC)Task: cost research v1DE · awaiting tech-ref reviewDoneTemplate adaptation✓ PM · acceptedComms strategy v1✓ PM · acceptedTwo acceptance layers — Definition of Done (internal) + Acceptance Criteria (client).DoD speeds up internal review; AC ensures the client gets exactly what the contract said they would.
09 — Foundation Established

What this project still gives me, years later.

Looking back, the value of this engagement wasn't only the platform I shipped — it was the planning model I built doing it. The patterns below are still the ones I anchor delivery work to today.

Carried forward

1. Three-level backlog as the universal translator

The Deliverable / Activity / Task hierarchy works because it speaks to three different audiences at once. The Deliverable layer is the contract — what the client signs off. The Activity layer is the conversation — where scope and effort actually get negotiated with the team. The Task layer is the sprint — what gets done this week. Most delivery friction I've seen since comes from teams collapsing two of these into one.

Carried forward

2. First sprint as a calibration, not a commitment

Sprint 1 was deliberately planned without slack — not because I expected the team to hit it perfectly, but because I needed to measure the gap. Velocity is empirical, and you only get a real number by running a tight first iteration and watching what actually lands. Every engagement I've led since starts the same way.

Carried forward

3. Quarterly milestones that a non-technical stakeholder can verify

“Velocity is up” is not a milestone. “Data is in the raw layer” is. Every quarter ended on something the sponsor could see or touch — a working dashboard, a deployed model, a signed report. The discipline of writing the milestone before designing the work has saved more programs from drift than any process artifact I've used since.

The platform was the deliverable. The planning model was the asset. Every delivery engagement I've led since has been a variation on the framework this project forced me to commit to paper.