01 /Gallery

Projects

What the wall is made of — how each piece is built, which shared platform elements it stands on, and how those elements connect. Walk the gallery itself on the home page.

← Walk the gallery

02 /At a glance

The wall in numbers

Projects

on the wall

Deployed

reachable right now

Distinct technologies

across every stack

Shared elements per project

4.1

of 8 available

Status mix

live5
beta3
experiment1

Most-used technologies

TypeScript8 projects
Next.js7 projects
PostgreSQL5 projects
Prisma4 projects
Redis3 projects
Zod3 projects
BullMQ2 projects
Playwright2 projects
Vitest2 projects
Anthropic API (behind an eval gate)1 project

03 /System

How the pieces connect

Experience

One Next.js app per product, behind a single reverse proxy.

Task Management7 shared elements

Smart Operations Dashboard3 shared elements

Testora4 shared elements

AI Evaluation Lab2 shared elements

EduMatch3 shared elements

Vionto Studio7 shared elements

TimelineAI4 shared elements

ResuMatch4 shared elements

AI Workbench3 shared elements

Shared platform

Built once in packages/*, consumed by every app above.

Design system@asafarim/uiTokens, moods, and brand components.9 of 9 projects

Identity@asafarim/authAuth.js config, SSO cookie, RBAC helpers.7 of 9 projects

AI boundary@asafarim/appbuilder-aiServer-only provider boundary.5 of 9 projects

Job runnerBullMQ workerDurable background pipelines.2 of 9 projects

State

Where things are persisted, queued, and stored.

Platform database@asafarim/dbShared Prisma schema on PostgreSQL.5 of 9 projects

Isolated databaseDrizzle + PostgreSQLApp-private schema, separate instance.3 of 9 projects

Object storage@asafarim/storageS3-compatible media and exports.3 of 9 projects

Queue / cacheRedisJob queues and rate limiting.3 of 9 projects

7 of 9 projects are deployed and reachable today. Every app in the top tier authenticates through the same identity package and renders with the same design system — the reason a new product starts as a page, not a stack.

04 /Coverage

Who builds on what

Shared platform elements used by each project. A filled cell means the project builds on that element directly.
Project	Design system	Identity	AI boundary	Job runner	Platform database	Isolated database	Object storage	Queue / cache	Total
Task Management	Task Management uses Design system	Task Management uses Identity	Task Management uses AI boundary	Task Management uses Job runner	Task Management does not use Platform database	Task Management uses Isolated database	Task Management uses Object storage	Task Management uses Queue / cache	7
Smart Operations Dashboard	Smart Operations Dashboard uses Design system	Smart Operations Dashboard uses Identity	Smart Operations Dashboard does not use AI boundary	Smart Operations Dashboard does not use Job runner	Smart Operations Dashboard uses Platform database	Smart Operations Dashboard does not use Isolated database	Smart Operations Dashboard does not use Object storage	Smart Operations Dashboard does not use Queue / cache	3
Testora	Testora uses Design system	Testora uses Identity	Testora uses AI boundary	Testora does not use Job runner	Testora does not use Platform database	Testora uses Isolated database	Testora does not use Object storage	Testora does not use Queue / cache	4
AI Evaluation Lab	AI Evaluation Lab uses Design system	AI Evaluation Lab does not use Identity	AI Evaluation Lab uses AI boundary	AI Evaluation Lab does not use Job runner	AI Evaluation Lab does not use Platform database	AI Evaluation Lab does not use Isolated database	AI Evaluation Lab does not use Object storage	AI Evaluation Lab does not use Queue / cache	2
EduMatch	EduMatch uses Design system	EduMatch uses Identity	EduMatch does not use AI boundary	EduMatch does not use Job runner	EduMatch uses Platform database	EduMatch does not use Isolated database	EduMatch does not use Object storage	EduMatch does not use Queue / cache	3
Vionto Studio	Vionto Studio uses Design system	Vionto Studio uses Identity	Vionto Studio uses AI boundary	Vionto Studio uses Job runner	Vionto Studio uses Platform database	Vionto Studio does not use Isolated database	Vionto Studio uses Object storage	Vionto Studio uses Queue / cache	7
TimelineAI	TimelineAI uses Design system	TimelineAI uses Identity	TimelineAI does not use AI boundary	TimelineAI does not use Job runner	TimelineAI uses Platform database	TimelineAI does not use Isolated database	TimelineAI does not use Object storage	TimelineAI uses Queue / cache	4
ResuMatch	ResuMatch uses Design system	ResuMatch uses Identity	ResuMatch does not use AI boundary	ResuMatch does not use Job runner	ResuMatch does not use Platform database	ResuMatch uses Isolated database	ResuMatch uses Object storage	ResuMatch does not use Queue / cache	4
AI Workbench	AI Workbench uses Design system	AI Workbench does not use Identity	AI Workbench uses AI boundary	AI Workbench does not use Job runner	AI Workbench uses Platform database	AI Workbench does not use Isolated database	AI Workbench does not use Object storage	AI Workbench does not use Queue / cache	3

05 /Detail

Project by project

Task Management

Beta

An AI-native work operating system that turns scattered intent into explainable, outcome-linked execution. Fifteen milestones shipped as a reviewed PR series; the creative & useful copilot epic is underway.

Details →

Next.jsTypeScriptPostgreSQLRedisBullMQOpenAPI

M00–M15 shipped as a reviewed PR series: multi-tenant work graph, versioned REST API with OpenAPI, idempotency and optimistic concurrency.
The fast task experience — subtasks, dependencies, estimates, labels, custom fields, recurrence and templates across Inbox, List, Board, Calendar and Timeline — with no AI required.
Human-approved AI proposals convert briefs and notes into tasks, dependencies, owners, and delivery plans.
Next: the creative & useful copilot epic (#244).

Smart Operations Dashboard

Beta

Operations KPI dashboard showcase with real-time views and reporting.

Details →

ReactTypeScriptRechartsPostgreSQL

KPI tiles and trend views composed from the shared design-system primitives.
Reporting queries read the platform schema directly — no separate warehouse.

Testora

Live

Live end-to-end test automation: author requirements, suites, fixtures and cases, run them in a real headless TestCafe browser with streamed progress, and triage failures with screenshots and AI-drafted issues.

Details →Open live ↗

Next.jsTypeScriptTestCafeDrizzlePostgreSQL

Full authoring model — functional requirements, test suites, fixtures, and cases — persisted in Testora's own Drizzle database, isolated from the shared platform schema.
Signed-in users drive real headless TestCafe runs against per-app local or remote targets, with progress streamed over SSE.
Repeat runs per case and a per-fixture flaky flag that adds retries.
Failure screenshots and AI-drafted issue reports; next up: the autonomous quality loop with TasksAI (#269).

AI Evaluation Lab

Live

A provider-neutral, fixture-mode AI benchmark: versioned prompts and synthetic datasets scored for correctness, groundedness, format compliance, latency, cost, and safety — reproducibly, with no API keys.

Details →Open live ↗

TypeScriptZodVitestProvider-neutral adapters

Versioned prompts and synthetic datasets — no API keys, no employer data.
Scores correctness, groundedness, format compliance, latency, cost, safety.
Models appear as provider-neutral aliases so runs stay comparable.
Latency and cost are representative fixtures, never live measurements.

EduMatch

Live

An explainable tutor-matching benchmark: synthetic students and tutors, a transparent weighted-factor engine you can adjust live, and fairness/stability checks.

Details →Open live ↗

Next.jsTypeScriptPrismaPostgreSQL

Synthetic students and tutors — no real people, bookings, or payments.
Transparent weighted-factor engine you can adjust live in the browser.
Fairness and stability checks run against the committed fixtures.

Vionto Studio

Live

A transparent AI media-pipeline benchmark: a schema-validated brief-to-render pipeline with approval-gated retry, seeded stage failures, and cost estimation — no live providers, no real media.

Details →Open live ↗

Next.jsBullMQFFmpegPrismaS3Redis

Schema-validated brief-to-render pipeline with approval-gated retry.
Seeded stage failures make the recovery path observable, not hypothetical.
Render work runs on a durable BullMQ worker, not in the request cycle.
Cost estimation per render, with no live providers and no real media.

TimelineAI

Live

Turn a list of events into a polished visual timeline: ten layouts over one content model — vertical, zigzag, circular, roadmap, Gantt, calendar board, and more — with live preview, PNG/JPG/PDF export, and a public gallery. No account needed to try it.

Details →Open live ↗

Next.jsTypeScriptPrismaPuppeteerPlaywrightRedis

One content model, ten layouts — switching layout never touches event data.
Live editor preview renders through the exact component the public page uses.
PNG / JPG / PDF export via headless Chromium against the same renderer.
Guests can build and submit a timeline without an account; admins moderate.

ResuMatch

Beta

An AI CV-tailoring showcase: paste a job posting URL, and AI rewords and reprioritizes your confirmed profile toward it — never inventing an employer, a date, a degree, or a skill you did not list — then download it as a PDF.

Details →Open live ↗

Next.jsTypeScriptPrismaPostgreSQLZodS3-compatible storage

CV-derived profiles are editable, explicitly confirmed, and kept as immutable versions.
AI rewrites are structurally constrained: facts are copied from the profile in code, never taken from model output.
One job URL is fetched per candidate action — no job-board aggregation, no licensing dependency.
The public deployment is a non-commercial portfolio MVP, not a professional career or legal service.

AI Workbench

Experiment

Three narrow public AI tools — requirements to test plan, notes to action plan, text to cited timeline — where every item shows where it came from, the server removes what the input doesn't support, and a fixture-first eval gate decides when a tool may go live.

Details →Open live ↗

Next.jsTypeScriptZodAnthropic API (behind an eval gate)VitestPlaywright

Every item is labelled: quoted from the input, inferred with its assumption, or a suggestion.
Server-side checks remove fake test results, invented owners and deadlines, and false date precision.
A fixture-first eval gate in CI blocks any tool from going live until it passes.
Experimental today: the examples run; live generation stays off until each tool's gate and launch checks pass.