Work

Selected systems, experiments and engineering decisions.

Not every repository deserves equal weight. These are the projects that best show how I design systems, evaluate models and make technical trade-offs.

Collaborative repositories are labelled explicitly. Academic results are presented as academic results.

AI systems · product engineering · 2026

CareerOS

A safety-gated AI application platform where model workers can propose, but deterministic software owns workflow state, authorization and auditability.

RoleProduct architecture and implementation

Core designFail-closedSubmission authorization is issued only by a deterministic gate.

Case study →thestcasa/carrerOS ↗

CareerOS architecture showing candidate configuration and jobs flowing through an orchestrator, AI workers, a deterministic submission gate and audit archive.

LLM research · reproduction · domain adaptation · 2026

TISER Temporal Reasoning

A reproduction of ACL 2025 TISER on Qwen2.5-7B, extended with a context-memory conflict probe and a professional-tennis temporal reasoning adaptation study.

RoleCollaborative research; contributed tennis-domain adaptation and integration of the main / tennis training-evaluation workflows

Collaborative project. Repository hosted by a collaborator; ownership is not presented as solo work.

Full-test reproduction0.949 F1Macro F1 across the five in-domain splits; macro EM 0.878 over 22,014 test examples.

Case study →thestbobo/tiser_temporal_reasoning_extension ↗

TISER pipeline showing structured temporal traces, LoRA fine-tuning, baseline evaluation, context-memory conflict analysis and tennis domain adaptation.

tabular ML · feature engineering · imbalanced classification · 2026

Credit Default Prediction

An end-to-end credit-risk classification workflow that turns repayment behaviour into robust features, tunes the operating threshold and finished first on the course leaderboard.

RoleModeling, feature engineering and reproducible experiment pipeline

Course leaderboard1st placePublic Macro F1 0.721 with a CatBoost configuration and decision threshold 0.328.

Case study →thestcasa/chalenge_credit_default_prediction ↗

Credit default machine-learning pipeline from raw behaviour through feature engineering, stratified cross-validation, CatBoost and threshold selection to final predictions.

NLP · data quality · evaluation discipline · 2026

News Classification Pipeline

A leakage-aware multiclass news classifier built around schema validation, duplicate analysis, drift checks, sparse text features and stratified cross-validation.

RolePipeline design, data-quality controls and evaluation workflow

Engineering focusLeakage-awareEvaluation discipline and data-quality controls are first-class stages of the pipeline.

Case study →thestcasa/news-classification-pipeline ↗

News classification pipeline showing schema and leakage checks, sparse text and metadata features, cross-validation, model selection and reporting.

time-aware ML · uncertainty · portfolio research · 2026

Regime Residual Portfolio

A residual-alpha forecasting and portfolio-construction research pipeline built around walk-forward evaluation, validation embargoes, MC Dropout uncertainty and transaction-cost awareness.

RoleResearch pipeline design and implementation

Research disciplineWalk-forwardTemporal leakage controls and explicit baselines are part of the evaluation design.

Case study →thestcasa/RegimeResidualPortfolio ↗

Walk-forward machine-learning diagram for regime residual portfolio research, with rolling training windows, uncertainty filtering and cost-aware portfolio construction.

deep learning · federated learning · efficient adaptation · 2025

Federated Vision Transformers

A collaborative research codebase comparing centralized and federated Vision Transformer training on CIFAR-100 with DINO backbones, LoRA and TaLoS-style model editing.

RoleCollaborative implementation and experiment configuration

Collaborative project. Repository hosted by a collaborator; contributions are presented as team work.

Research surfaceCentral + federatedThe same codebase supports centralized and client-server experiments with multiple fine-tuning strategies.

Case study →thestbobo/DLML_FL_project ↗

Federated learning diagram showing several non-IID clients fine-tuning Vision Transformer models and sending updates to a central aggregation server.

Smaller coursework and technical explorations.

2026Network Dynamics & Learning

Executable Python work across flows, centrality, Wardrop equilibria, random walks, opinion dynamics, epidemic models and network games.

GitHub ↗