01
TCAF defines the rules for AI-assisted work
It makes task boundaries, preservation of working code, review, verification, and project continuity reusable across tools and codebases.
Projects
Three public projects showing how I approach software from implementation to project structure.
TCAF documents a controlled way to use AI in software development. Software Delivery Planner makes planning and engineering decisions visible, while Travel AI Flight Optimizer shows end-to-end product work with live providers and deterministic application logic.
Featured project · Public beta v0.3.3
Task-Contract AI Development Framework — TCAF
An open framework for professional software development with AI that can be used across different tools and models. TCAF keeps product decisions, architecture, implementation strategy, code responsibility, quality, and final acceptance with developers and teams.
AI tools receive only selected work through explicit Task Contracts. The framework defines what may change, protects working code from unnecessary rewrites, keeps project context available, records what has actually been verified, and requires developer review before acceptance.
Project profile
Type
Open-source development framework
Goal
Make AI-assisted changes bounded, reviewable, verifiable, and easier to integrate into normal software work.
Status
Public pre-1.0 beta
Verified integrations
Native Codex integration and a documented Cline manual workflow, both with known limitations recorded.
Versioned runtime and CLI
Universal Run Protocol
Task Contracts and developer review
Rules to preserve working code
Verified Codex integration
Documented Cline workflow
Project setup and task flows
Bilingual public documentation
Public application project · Work in progress
Software Delivery Planner
A full-stack portfolio application for planning and monitoring a small software project. It combines project scope, feature tracking, architecture, estimates, risks, KPIs, and supporting documentation in one intentionally lightweight product.
The repository contains working project management and feature flows, shared frontend/backend contracts, automated tests, OpenAPI documentation, documented architecture and MVP scope, and a public backlog. The product is still in progress and its current status is stated openly.
Current implementation
React, TypeScript, and Fastify workspace
Working project management flow
Working feature backend flow
Shared frontend/backend contracts
Automated API and UI tests
OpenAPI / Swagger documentation
Documented architecture and MVP scope
Public backlog and planning material
Public application project · MVP complete
Travel AI Flight Optimizer
A local-first travel planning application that combines live flight discovery, deterministic shortlist ranking, home-to-airport transport planning, source-backed destination research, and exportable reports in one end-to-end user flow.
The MVP uses Kiwi MCP for live flight offers, keeps ranking and compatibility decisions deterministic, and uses AI only where web information needs to be organized or summarized. It supports exact or flexible dates, multiple airports, open-jaw combinations, reusable transport data, Destination Insights, Kiwi booking links, and standalone AngeliniLabs-branded HTML reports.
Current implementation
React, TypeScript, Vite, and Express
Live flight discovery through Kiwi MCP
Exact/flexible dates and multi-airport search
Open-jaw round-trip combinations
Deterministic filtering and shortlist ranking
Reusable home-to-airport transport rules
Destination research backed by sources
Standalone branded HTML reports
How they connect
Three projects, three views of the same way of working.
Each project can be evaluated on its own, but together they show development, project planning, technical decision-making, testing, integrations, and controlled use of AI.
02
Software Delivery Planner makes planning visible
The application connects frontend and backend development with scope, backlog, estimates, risks, tests, architecture, and project documentation.
03
Travel AI Flight Optimizer applies the approach to a live product
The travel MVP combines external providers, web research, selective AI support, deterministic ranking, transport rules, and iterative product development without delegating core product decisions to the model.
What this portfolio demonstrates
Development, technical planning, and project coordination in the same body of work.
The objective is not to show isolated code samples. It is to make the decisions, constraints, trade-offs, quality controls, and evolution of the projects visible.
TCAF demonstrates framework design, runtime architecture, reusable development rules, validation, tool-integration constraints, documentation, and release discipline.
Software Delivery Planner demonstrates React and TypeScript frontend work, Fastify APIs, shared contracts, validation, automated testing, OpenAPI, and project material such as scope, backlog, risks, architecture, and estimates.
Travel AI Flight Optimizer demonstrates provider integration, deterministic multi-criteria ranking, transport compatibility, web research, selective AI-assisted synthesis, React/TypeScript product work, Express APIs, automated tests, and local-first reporting.
Together they show a profile that can move between implementation, technical decisions, project planning, coordination, and responsibility for software execution.
Roadmap
What comes next.
The public projects will continue to evolve incrementally, with current status and limitations kept visible.
TCAF case studies
Public examples showing Task Contracts, controlled implementation, verification results, project updates, and lessons learned across different kinds of software work.
TCAF integrations
Support for additional development tools and models, added only after their practical limitations and integration behaviour have been tested.
Application project evolution
Software Delivery Planner and Travel AI Flight Optimizer will continue to evolve incrementally, with new capabilities added only when they are coherent with each project's real objectives and current state.
Inspect the work
Repositories and documentation are public.
The projects expose their implementation, documentation, current status, and known limitations so the work can be inspected directly rather than judged only from a description.