>_ Jorge Uc / projects← All projects
~/projects/pokeponche // END-TO-END CASE STUDY
◆ Real-world impactAndroid + MLPuebla, Mexico

From one trainer to a thriving community.

PokePonche de Frutas started with one organizer and a handful of players. With limited time, volunteers and donated prizes, I built an Android event-planning tool and a machine-learning forecasting workflow to help the community organize meetups, understand attendance and make more informed resource decisions.

Large PokePonche community meetup in Puebla, Mexico
COMMUNITY / PUEBLA2026 · REAL EVENT PHOTO
EARLY COMMUNITY~11players in a 2024 photo
LARGE MEETUPS300+participants reported
EVENT DATASET548meetups analyzed
DEMO MODEL MAE6.75check-ins per event
01 / PROJECT CONTEXT

When community growth depends on small decisions.

At the beginning, a few donated plushies, figures and small gifts had to be allocated carefully. Every event also required artwork, descriptions, invitations and organizer time.

01 / TIME

Repetitive event creation

Preparing graphics and copying details into Campfire consumed time that could have been spent supporting participants.

02 / DATA

Scattered participation records

Useful attendance and RSVP information existed, but it needed consistent collection and organization before analysis.

03 / RESOURCES

Limited prizes and volunteers

Choosing which events deserved greater preparation was an important decision for a growing community.

02 / CONNECTED SOLUTION

Two products. One decision-making loop.

An Android workflow handles event operations and historical exports. A Python pipeline transforms those records into predictive features for meetup planning.

> community.data_loopCONCEPTUAL WORKFLOW
01 / OPERATIONSAndroid meetup creatorArtwork · Titles · Sharing · Records
02 / COLLECTIONCampfire attendanceCheck-ins · RSVP · CSV / ZIP exports
03 / PREDICTIONCatBoost forecastingHistorical behavior · Event signals
04 / ACTIONSmarter event planningEffort · Supplies · Prizes · Recruitment

Campfire events are created manually; the Android app prepares their content and stores the event links. Attendance retrieval relies on third-party Campfire Tools.

03 / ANDROID APPLICATION

Making meetup preparation repeatable.

Built with Kotlin and Jetpack Compose, the companion app streamlines artwork, meetup information, invitations, attendance records and prize raffles.

FROM DESIGN TO INVITATION

One workflow instead of separate manual tasks.

The application generates event images with up to three Pokémon, composes editable titles and descriptions, stores Campfire links, and prepares invitations for sharing.

Organizers can later download check-ins, inspect attendance, export structured files and run raffles from saved participant lists.

KotlinJetpack ComposeLocal JSONAndroid ShareSheetCampfire Tools
Real Android screenshot showing meetup artwork generation
REAL APP SCREEN / ARTWORK GENERATOR
04 / DATA ENGINEERING

From event history to a clean dataset.

Organized attendance data became the foundation for historical trends, community activity measures and predictive features.

MEETUPS548event-level observations
INTERACTION RECORDS33,848used to build historical features
UNIQUE USERS5,131pseudonymized
CHECK-INS23,032across the dataset
DATASET DESIGN

One row per event, with historical behavior derived from participation records.

The model learns from 548 meetups, not 33,848 independent training examples. Individual check-in histories are aggregated into event-level features such as active members, recent turnout, repeat participation and previous conversion rates.

Event dates, categories and bonuses are combined with prior community activity. Historical features are calculated only from earlier dates to respect the sequence of events.

PythonpandasTemporal featuresAggregationPseudonymization
05 / MACHINE LEARNING

Predicting participation, not just counting RSVPs.

The published demonstration reconstructs a CatBoost-based prediction workflow used to support meetup planning and allocation of limited resources.

MODEL / CATBOOST REGRESSION

A historical conversion approach.

The selected model learns the expected conversion of interested participants into actual Campfire check-ins. The estimated conversion rate is then combined with the observed interest count to produce an event-level attendance estimate.

Event + historical signals→Estimated conversion→Predicted check-ins

This model forecasts event-level check-ins, not the identities or behavior of specific players.

DEMO MODEL
FEATURES4641 numerical / 5 categorical
OPTUNA TRIALS20Temporal validation
FINAL TEST SET81chronologically held-out events
01 / CALENDAR

Timing matters

Seasonality, day of week, event sequencing and timing of major global Pokémon GO events.

02 / COMMUNITY

Recent history matters

Rolling attendance, user activity windows, repeat participation and historical conversion.

03 / EVENT

Context matters

Meetup category, event phase, available bonuses, Pokémon-related signals and interest count.

06 / DEMONSTRATION RESULTS

Model quality, shown with real plots.

Public, reproducible model results from the 81-event chronological test set. Figures are taken from the project itself.

MAE6.75check-ins
MEDIAN ABS. ERROR5.43check-ins
R²0.972test evaluation
WAPE13.65%test evaluation
01 / TIME SERIESActual vs. predicted check-ins
All 81 held-out events, presented in chronological order.
02 / FITPredictions vs. actuals
The dashed diagonal shows the ideal prediction.
03 / ERRORSAbsolute error distribution
Half of absolute errors were 5.43 check-ins or lower.
07 / REAL-WORLD OUTCOMES

From a handful of players to events with hundreds.

The strongest evidence of this project's relevance is the community it served. The tools supported planning and operations alongside the sustained work of volunteers, organizers and participants.

Early PokePonche meetup in Puebla in 2024 with a small group of participants
2024 / THE BEGINNING~11 playersA small local meetup
Large PokePonche group at a recent event in Puebla in 2026
2026 / A GROWING COMMUNITY300+ participantsA large community gathering
GO FEST / JULY 11–12, 2026

GO Fest 2026 Puebla

PokePonche helped organize a major Pokémon GO community gathering in Puebla as part of GO Fest 2026. The photo shown here features the event's PokePonche branding and an energetic crowd of attendees.

The community's growth reflects the efforts of its organizers and members; software and data were tools that helped coordinate the work, not the sole cause of that growth.

PokePonche Puebla photo from Pokémon GO Fest 2026, with event branding and attendees
GO FEST 2026 / PUEBLA, MEXICO
WHY IT MATTERS

The value was not just predicting a number. It was helping people decide what to do with limited time and resources.

Attendance estimates informed decisions about when to invest more effort, prepare donated prizes and focus recruitment activities. Operational data and community experience worked together.

08 / PERSONAL RETROSPECTIVE

What I learned from building PokePonche.

Keeping a real community running changed how I think about engineering: good models depend on dependable data collection, clear workflows and actionable decisions.

01 / PRODUCT ENGINEERING

Automation creates room for people

Removing repetitive graphic and announcement tasks gave organizers a more consistent workflow and more time to focus on the meetup itself.

AndroidKotlinUX
02 / DATA ENGINEERING

Reliable records are a product feature

Transforming meetup links, check-ins and participant histories into structured exports taught me to design collection and analysis together.

ETLCSVData quality
03 / APPLIED ML

Time-aware evaluation changes the model

Historical feature windows, chronological splits and conversion modeling mattered because meetup attendance evolves along with the community.

CatBoostOptunaFeature engineering
04 / DECISION SUPPORT

Prediction is not the final step

A useful estimate becomes valuable when it helps determine where to place volunteers, gifts and preparation effort, while still leaving room for human judgment.

AnalyticsOperationsResource planning
KEY TAKEAWAY

Build for a real problem, collect data responsibly, and turn predictions into better choices.

This project combined community organizing, Android development and data science into one evolving workflow, shaped by the people who actually used it.

09 / SOURCE & ATTRIBUTION

Explore the engineering.

The two codebases can be presented separately on GitHub while this page tells the complete end-to-end story.

ANDROID / KOTLIN

Campfire Meetup Creator

Artwork generation, meetup preparation, sharing, check-in exports and raffles. Android 10+.

Visit GitHub profile ↗
PYTHON / MACHINE LEARNING

PokePonche Check-In Predictor

Reproducible CatBoost demonstration with pseudonymized data, model artifacts and evaluation plots.

Visit GitHub profile ↗
INDEPENDENT COMMUNITY PROJECT

Pokémon, Pokémon GO, Campfire and related names, character art and marks belong to their respective owners. This independent project is not affiliated with, sponsored by, or endorsed by Niantic, Scopely, The Pokémon Company or Nintendo. Attendance data shared publicly are minimized and pseudonymized; participant identities and internal operational records are not published.