Skip to content
Almathal Handbook

MVP Scope

Accepted

This document defines what is in scope and what is explicitly out of scope for the Almathal MVP. The MVP is the first shippable version that proves the architecture and supports paying enterprise customers on a limited workload.

MVP Goals

  1. Prove the retrieval-first thesis with measurable economics: archetype-matched generations should cost meaningfully less than equivalent pure-LLM output.
  2. Establish trust guarantees that pure-LLM tools structurally cannot offer: per-component provenance, license verification, CVE scanning, audit trails.
  3. Ship two Archetypes end-to-end through the full pipeline, with deterministic gates and bounded LLM stitching.
  4. Build the curation infrastructure so the library can grow with the team rather than blocking on it.

What Ships in MVP

Archetypes (2)

  1. RAG Chatbot over private corpus (ADR-0022)
  2. Internal CRUD Admin Tool (ADR-0023)

Each Archetype takes generation end-to-end: Spec Resolution → Compatibility Validation → Build Approval → Seam Validation → Scaffolding → Stitching → Build Verification → Output Review → Delivery.

Component Library

Approximately 150 Adapters at MVP, across four slices:

  • Java slice (~50 adapters): Spring Boot, JPA, Security, Kafka, JHipster integration, Micrometer, etc.
  • TypeScript/JS slice (~35 adapters): React, Next.js, Tailwind, shadcn/ui, TanStack Query, Zod, etc.
  • Python slice (~30 adapters): FastAPI, pandas, polars, pydantic, SQLAlchemy, etc.
  • AI/LLM slice (~25 adapters): Anthropic SDK, OpenAI SDK, pgvector, LangChain, sentence-transformers, etc.
  • Cross-cutting infrastructure (~15 containers/services): PostgreSQL, Valkey, Kafka, MinIO, Keycloak, Prometheus, Grafana, Nginx, OpenAPI tooling.

Detailed inventory at Component Library Seed.

Module Library

Approximately 5–8 Modules at MVP, covering only what the two MVP Archetypes need:

  • module:auth/jwt-with-rbac — authentication for the CRUD admin Archetype
  • module:ingestion/standard-corpus — document ingestion for the RAG Archetype
  • module:rag/chat-with-citations — core RAG loop with citation rendering
  • module:observability/standard-stack — Prometheus + Grafana + structured logging
  • module:audit-log/event-stream — audit trail emission for compliance
  • Possibly 2–3 more as Archetype design evolves

Composer

Full implementation of all pipeline stages:

  • Spec Resolution (Resolver subcomponent) with hybrid form + conversation input
  • Compatibility Validator (deterministic matrix walk)
  • Briefing Generator (LLM summary for Build Approval)
  • Seam Validator (static analysis)
  • Scaffolder Runner (JHipster, Spring Initializr, create-t3-app)
  • Build Validator (compile, lint, test runners)
  • Handoff to Stitcher

Stitcher

Bounded LLM layer with:

  • Prompt templates per Seam type
  • Multi-model support (configurable per stage)
  • Retry logic on Build Verification failure
  • AI Reviewer (using a different model from the Stitcher)
  • Token accounting and cost reporting per generation

Trust Infrastructure

  • Central Manifest Registry with full audit trail
  • License verification (SPDX-based)
  • CVE scanning integrated into the Curation Pipeline
  • SBOM emission with each generated app
  • Build records (UUIDs, slug-at-build-time, version locks) attached to every generation
  • Customer-accessible audit trail

Curation Pipeline

  • Discovery automation for upstream releases and CVEs
  • Stage 1 (LLM-drafted, human-directed) Manifest authoring workflow
  • Stage 2 (LLM-drafted, human-reviewed) workflow for non-AI slices
  • Mechanical evaluation checks
  • Weekly patch / monthly minor release cadence
  • Decision log

Customer Surface

  • Web UI for Archetype selection, form-based Variation Point input, conversation refinement, Build Approval
  • Generated app delivery (code, containers, tests, audit trail)
  • Saved Spec templates (per ADR-0014)
  • Audit trail viewer
  • Release notes and changelog

What Is Explicitly Out of Scope for MVP

The following are real future work but not in MVP:

Archetypes deferred

  • Microservices Archetypes (deferred to v1 per ADR-0015)
  • E-commerce, marketplace, POS, billing Archetypes
  • AI agent / multi-tool workflow Archetypes
  • Document processing pipeline Archetypes
  • Embedded analytics Archetypes
  • Compliance reporting tools

Component Library deferred

  • Cloud-vendor-specific SDKs (AWS, GCP, Azure) — added in v1.5 “cloud pack”
  • Heavyweight enterprise integrations (IBM MQ, Oracle DB, SAP) — added on paying-customer demand
  • Mobile frontend (React Native, Flutter) — out of enterprise web focus
  • Additional language slices (Go, Rust, C#, .NET) — post-v1 expansion
  • Legacy framework support (Spring 5, React 17) — only current stable versions in MVP

Module Library deferred

  • Payments Module (Stripe, Adyen wrappers) — v1
  • Billing Module (Lago wrapper or platform-composed) — v1
  • Email/notification Modules — v1
  • File upload and storage Modules — v1
  • Search Modules (full-text search beyond what RAG covers) — v1
  • Workflow / approval engine Modules — v1+

Pipeline features deferred

  • Exploratory archetype path (replaces pure-LLM fallback) — v1
  • Cross-build re-generation with customization preservation — v1
  • Multi-app orchestration — v1 (alongside microservices Archetypes)
  • Live editing of generated apps within the platform — possibly never (out of focus)

Trust infrastructure deferred

  • Optional verified artifact mirror (per ADR-0017) — v1 enterprise tier
  • On-prem / air-gapped deployment — v1 enterprise tier
  • Customer-specific Manifest extensions — v1+
  • SOC 2 Type II audit (Type I is acceptable for MVP launch) — v1
  • HIPAA, PCI-DSS specific compliance packs — v1+

Customer surface deferred

  • Marketplace for community Adapters / Modules — post-v1
  • Multi-user team collaboration on a Spec — v1
  • CI/CD integration for auto-regeneration — v1
  • API access (machine-driven generations) — v1
  • IDE plugins — v1+

MVP Success Criteria

The MVP is judged on:

  1. Both Archetypes produce working generated apps. Compile, deploy, smoke tests pass.
  2. Token cost ratio. Archetype-matched generations cost at least 2× less than comparable pure-LLM tools producing the same app (target: 5–10×).
  3. Provenance completeness. Every Adapter and Module in every generated app has full provenance recorded in the build’s audit trail.
  4. Build reproducibility. Re-running the same Spec produces functionally equivalent output (mechanical verification passes identically).
  5. Curation cadence. The Curation Pipeline can sustain weekly patch releases without falling behind on upstream CVEs.
  6. First enterprise pilot. At least one paying enterprise customer running real generated apps in production by MVP+90 days.

Timeline

A rough estimate, contingent on team size and AI-assisted-development efficiency:

PhaseDurationOutput
Foundation4–6 weeksComponent Library MVP corpus (Stage 1 authoring), Manifest registry running, schema validation in place
Composer4–6 weeksPipeline stages implemented end-to-end for one Archetype (CRUD admin first, as the simpler verification surface)
Stitcher4–6 weeksSeam-driven generation, multi-model support, AI Reviewer, verification harness
Second Archetype3–4 weeksRAG chatbot Archetype on top of foundations
Polish & Trust3–4 weeksAudit trail, license/security infrastructure, customer UI refinement
Enterprise readiness3–4 weeksSSO, deployment options, SOC 2 Type I preparation

Total: roughly 5–7 months from start of focused implementation to MVP-ready, with a small team of 2–3 senior engineers using AI-assisted development aggressively.