Skip to content
Tenzok
Home
Services
Student Projects
BlogAboutContact
All student projects

5 scoped briefs

3 mini · 2 major

Data Analytics & Business Intelligence

This is the consumption end of the data stack: SQL that answers a question, a model that makes the answer repeatable, and a chart only at the very end. Data engineering owns whether the data arrives — on time, uncorrupted; this domain owns what it means once it is here: the grain, the metric definition, the segment cut, the confidence interval. Everyone has a dashboard, so a project only counts if it is defensible — student work fails here by being decorative: a Power BI page built straight on a flat CSV, with no fact grain, no baseline to beat, and a conclusion the data does not actually support.

Advanced SQLPostgreSQLDuckDBBigQuerydbtPower BITableaustatsmodels / statsforecastMetabase
Start a projectDiscuss a custom brief

The standard

Build the system. Understand every trade-off.

Every brief ends with working software, documented decisions, and evidence you can present—not a folder of code you cannot explain.

Two audiences, one engineering standard

Academic depth meets production discipline.

For students

Pick a brief below, or bring the problem statement your department handed you. We architect it with you, review every commit, deploy it to a real URL, and rehearse the viva until no question surprises you.

See student support

For companies

Semantic and dimensional modelling in dbt, metric definitions that stop two dashboards disagreeing about revenue, BI builds and migrations in Power BI or Tableau, and classical forecasting, experiment design and readouts — the layer above ingestion, where a number has to resolve to one meaning.

Discuss a company build

Mini projects

Focused scope. Real engineering.

Three to four weeks. Narrow enough to finish, deep enough that you learn the decision that actually matters.

3 briefs
Mini project3–4 weeks

Retail Star Schema & Power BI Report

Raw order CSVs modelled into a star schema in Postgres with a Power BI report on top — where the hard part is fixing the fact grain so measures stop double-counting across joins.

PostgreSQLSQLPower BIDAX

You walk away with

  • Star schema with the fact grain written down (one row = one order line) and conformed dimensions
  • A real date dimension table — fiscal periods, week-of-year, holiday flag — not a raw date column
  • Type-2 slowly-changing dimension on product price and category, with history preserved
  • Power BI report where measures are DAX, not visual-level aggregates
  • One-page insight memo: three findings, the SQL behind each, and a recommendation with its caveat stated
Mini project4 weeks

A/B Test Readout on a Public Experiment Dataset

A full experiment analysis — pre-registered plan, power simulation, peeking cost, effect size with an interval — ending in a written defence of what the result does and, more importantly, does not prove.

PythonNumPyscipystatsmodelsSQL

You walk away with

  • Analysis plan fixed before the outcome column is opened: primary metric, minimum detectable effect, sample size and power
  • Power simulation: an MDE curve showing the effect this sample could actually have detected, so a null result can be told apart from an underpowered one
  • Sanity checks: sample-ratio-mismatch test and an A/A comparison on pre-experiment data
  • Peeking cost quantified by simulation — the false-positive rate under repeated looks — with one sequential correction (alpha spending or an always-valid interval) applied to the same data
  • Effect size with a confidence interval, never a bare p-value; one pre-specified heterogeneous-treatment-effect check with its multiple-comparisons correction; and a decision memo — ship, do not ship, or re-run — stating what the result does not prove
Mini project4 weeks

Demand Forecast & Variance Dashboard

A monthly SKU demand forecast benchmarked against the seasonal-naive baseline it has to beat, published as a dashboard that tracks forecast error rather than just the forecast.

statsforecaststatsmodelspandasTableau Desktop (free student licence)PostgreSQL

You walk away with

  • Rolling-origin backtest — never a random train/test split on a time series
  • AutoARIMA and AutoETS (statsforecast) reported next to a seasonal-naive baseline on identical backtest windows, so a win is a real win
  • Error metrics per horizon: MAPE and MASE, because MAPE breaks on near-zero demand
  • Tableau dashboard connected to the Postgres warehouse, showing forecast, prediction interval, and last period's actual-vs-forecast error
  • A written list of the SKUs this model should not be trusted for — intermittent demand, short history, promo-driven spikes — and why

Major projects

Capstones you can defend.

Ten to twelve weeks. Architecture, trade-offs, failure modes, deployment, and evidence—the project that carries an interview.

2 briefs
Major project10–12 weeks

Semantic Layer: Governed Marts and a Metric Catalogue

Raw sources modelled through dbt into governed dimensional marts with a metric catalogue and a BI layer on top — where the hard part is the non-additive metrics that break the moment someone sums them.

dbtPostgreSQLBigQueryPower BIMetabaseSQL

You walk away with

  • Dimensional model at the mart layer: conformed dimensions reused across at least three fact tables, each fact's grain stated and enforced by a uniqueness test
  • A metric catalogue — for each of ~10 metrics: its definition in words, its exact SQL, its owner, and its known exclusions — so that 'revenue' resolves to exactly one definition across every mart and report in the project
  • Two hard metrics modelled and defended: one non-additive (a rate or ratio that cannot be summed across rows) and one requiring a point-in-time or period-over-period comparison
  • Reconciliation checks that catch two marts disagreeing on a shared metric, run on every model change, so a definition cannot drift in silence
  • BI layer built only on the marts — no report reading a source table directly — with row-level security on one dimension, plus a stakeholder summary a non-technical reader can act on without a follow-up question
Major project10–12 weeks

Product Analytics Investigation: Funnel, Retention, Causal Readout

A behavioural investigation over a large public event dataset — where the funnel leaks, which cohorts come back, what a past change actually did — ending in a recommendation list ranked by estimated impact and effort.

DuckDBSQLpandasstatsmodelsMetabase

You walk away with

  • Event model with the sessionisation rule defended: why a 30-minute inactivity window, and what it breaks
  • Funnel analysis with drop-off by step and by segment, including at least one segment cut (device, channel, or geo) reported with its finding stated honestly — including if the cut shows nothing
  • Cohort retention curves with a check that any movement is a retention change and not a cohort-mix change (Simpson's paradox)
  • Quasi-experimental readout on a change already visible in the data — a release, a policy change, a pricing shift — using interrupted time series or difference-in-differences, with the identifying assumptions written down and stress-tested (pre-period parallel trends, one placebo test). Pick the dataset for one.
  • An experiment design a team could actually run for the top recommendation — primary metric, MDE, sample size, run duration, guardrail metrics, pre-registered analysis plan — attached to a recommendation list ranked by estimated impact and effort, each with a stated confidence

Explore more

Other engineering domains

Search every brief
Python Full-StackJava & EnterpriseAI & LLM ApplicationsMachine LearningDeep LearningData EngineeringMobile AppsCloud & DevOpsCybersecurityIoT & EmbeddedJavaScript Full-StackVLSI & Chip DesignPower Electronics, EV & Energy SystemsSignal Processing & Wireless CommunicationsRobotics, Drones & Autonomous SystemsTest Automation & SDETComputer Networks & Protocol Simulation
Your next build starts here

Turn the idea into software people trust.

Bring us a product brief, a business problem, or a final-year project. We’ll turn it into a clear scope, a working build, and a handover you fully own.

Start Your ProjectSend your brief

Prefer email? info@tenzok.in

Tenzok

A product engineering studio for ambitious companies, founders, and students who want real, production-minded work.

info@tenzok.in

Company

HomeBlogAboutContactFAQ

Services

MentorshipStudent ProjectsCompany ServicesDigital MarketingLaunch Support

Project domains

Python Full-StackJava & EnterpriseAI & LLM ApplicationsMachine LearningExplore all 18 domainsRSS feed

© 2026 Tenzok. All rights reserved.

Obsession · Purpose · Excellence