za.
All work

HEALTH ANALYTICS / DECISION SUPPORT

StuntLytics

A child-health analytics workspace connecting regional evidence, risk mapping, and local screening workflows.

Role
Data preprocessing & system integration (hackathon)
Period
Original project: October 2025; workspace documented in 2026
Status
Prototype

Illustrative StuntLytics project preview; not a recording of validated clinical use

The problem

Regional child-health data is difficult to turn into a practical monitoring workflow. Teams need to compare conditions, review risk signals, and identify areas for follow-up.

Contribution & scope

Documented role: Data preprocessing & system integration (hackathon).

Project implementation

  • For the original APHacton project: a synthetic dataset of 100,000 family records using variables derived from BPS and SSGI references.
  • A Random Forest preprocessing and prediction workflow integrated with Streamlit and Elasticsearch in the original project.
  • A current workspace covering regional dashboards, risk maps, a Data Explorer, and family screening with 18 structured inputs.
  • An evidence layer that limits optional Gemini narration to aggregated information.

The approach

The current implementation combines a Next.js interface, FastAPI evidence services, Elasticsearch, and a local scikit-learn screening pipeline. Optional Gemini narration works from aggregated evidence. It develops the original hackathon concept into a broader analytics workspace. Regional Risk Intelligence is the West Java mapping module of StuntLytics, covering district-level monitoring, hotspots, and directional-flow signals; it is not counted as a separate project.

Architecture & workflow

  1. Regional dataset
  2. Elasticsearch
  3. FastAPI evidence service
  4. Local screening model
  5. Next.js workspace
  6. Monitoring + review

Technical challenges

  • Integrating preprocessing, model inference, indexed data, and the application interface.
  • Keeping screening signals distinct from clinical diagnosis.
  • Supporting local evidence summaries when optional external services are unavailable.

Outcomes

  • Delivered the original hackathon workflow for synthetic-data preparation and system integration.
  • The current repository brings dashboards, exploration, screening, and evidence summaries into one workspace.
  • The synthetic-data prototype and development-stage screening workflow are not presented as clinically validated performance.

Evidence & context

  • The current local screening pipeline accepts 18 structured inputs.

    Input count documented in the current repository. It is a feature count, not a clinical performance or accuracy claim.

    StuntLytics README — model and limitations

  • Prepared synthetic family data and integrated a Random Forest decision-support workflow for the original hackathon project.

    The CV identifies the original role as Data Preprocessing & System Integration. Synthetic data is not evidence of real-world or clinical validation.

    CV — project details

  • Regional Risk Intelligence is the regional mapping module of the StuntLytics workspace.

    The portfolio presents this map in the same application shell and links both entries to the same StuntLytics repository. It is not counted as a separate project or an independent validation result.

    Regional mapping image extracted from the owner's portfolio

Scope & limitations

  • The original hackathon workflow used synthetic family records; these are not clinical validation data.
  • The current workspace is development-stage decision support, not a diagnosis tool or a clinically validated screening service.
  • The preview video is illustrative project material, not evidence of clinical use or an independently verified interface recording.

What I learned

  • Dataset provenance matters: synthetic training records need to remain clearly distinguished from real-world validation.
  • Aggregating evidence before optional AI narration creates an explicit boundary around raw records.

Built with

  • Next.js
  • FastAPI
  • Elasticsearch
  • Scikit-learn
  • Gemini
  • Docker