Applied Analytics & Product Engineer

A score, a shortlist, or a learning path can look convincing before its assumptions are clear. I build the models and interfaces around those decisions, then show where the evidence comes from and where it stops. This portfolio spans exploration, implemented tools and concept products, with evaluation and honest maturity notes close to the work.

I built a credit-risk engineering study around PD scorecards, LGD/EAD, monitoring and reporting, then added synthetic-tested contracts for dated outcomes and reporting inputs. It is an engineering case study; historical model results and regulatory totals remain unverified.

An explainable location-evidence workstation that weighs conflicting signals, traces them through a graph and explains candidate postal-code rankings. The cases are synthetic; the reference catalog is not a set of live API integrations.

A learning experience for practicing better questions: a navigable 3D map of 147 concepts, 36 activity playbooks and six curated paths. The portal works as a learning prototype; its session runner and planner are incomplete.

Where Curiosity OS focuses on the learner's activities, VIZIER focuses on the architecture of knowledge: a compiler turns a 228-entry AI engineering syllabus and Markdown notes into a linked graph, searchable library and note reader. It is a personal learning system; the current curriculum snapshot does not mark its planned chapters as completed.

Independent studies in forecasting, credit risk, churn and reinforcement learning. The corrected forecasting companion compares rolling predictions with naive baselines; the corrected credit-risk and churn frameworks have synthetic software tests but no new real-data results.

A consumer food concept carried into an editorial storefront, sample product catalog, recipe collections and browser-local cart. It demonstrates brand and interaction design, not a live commerce operation or verified product claims.

• Make the decision path inspectable. LOC-IQ exposes evidence provenance; Therapy Matching OS separates eligibility rules from ranking; the credit-risk project gates interpretation on the data it needs.
• Treat evaluation as part of the build. The forecasting companion uses rolling comparisons against simple baselines. The CartPole companion reports variation across seeds and does not claim a reward-shaping win from a small observed difference.
• Design the whole journey. The projects connect research or content models to navigation, feedback and interaction, whether that is a learning path, a risk review, a daily reflection or a product concept.

Matching and everyday wellbeing: Therapy Matching OS is a seeded, rule-based discovery concept, not a clinical service; Runway 250 explores a daily-action loop against a seeded journey; Trelis Life explores small wins and reflection.

Knowledge and publishing: WARREN links 304 draft, unverified banking-risk atoms; Quant OS is a finance reading graph; Loud Recall is a public-note library.

Venture and operations concepts: FMCG Whitespace OS lays out product and channel hypotheses; Punarnava models a proposed recycling process in an interactive factory walkthrough. Both are concept studies, not operating businesses.

The question behind much of this work: what would someone need to see and understand before trusting a system's recommendation?