Personal projects

Built to understand, not to copy-paste.

Three projects, three layers of the same interest: financial software done properly — on-device, in the model, and in the audit trail.

iOS · SwiftUI + SwiftData

BudgetBook

A personal-finance app built to learn fundamentals properly: nine @Model entities (accounts, cards, loans, schedules, transactions, splits) with cascade-delete relationships, a LoanCalculator service implementing real EMI amortisation, and Swift Charts throughout. ~36 Swift files, feature-organised.

github.com/thegayankalinga/BudgetBookSwiftUiApp

budgetbookActive

▸9-entity SwiftData model with @Relationship cascade rules

▸EMI engine: E = P·r·(1+r)ⁿ / ((1+r)ⁿ − 1), maturity & schedules

▸Dashboard, budgets, credit cards, loans — Swift Charts

Applied ML · MSc research

SEE — Software Effort Estimation

MSc thesis turned running system: a hybrid LSTM–XGBoost model with feature-level fusion, trained on a proprietary FinTech delivery dataset, served through a FastAPI microservices backend and web frontend.

github.com/thegayankalinga/Software-Effort-Estimation-Model

see — software effort estimationResearch → product

▸23% RMSE improvement vs standalone XGBoost

▸40% improvement vs COCOMO baseline

▸BFF gateway · model service · user mgmt · effort calc, load-balanced

Agentic AI · in progress

Ledger

An auditable, agentic document-decisioning service for regulated finance — every decision traceable, every model judgement evidenced. Built in the open with a numbered build log, eval metrics and before/after tables.

Follow the build log →

ledgerBuild log live

▸Audit-trail-first architecture

▸Eval harness as a first-class artefact

▸Follows the publish-as-you-build cadence