Software Architect GPT-AI software architecture assistant

Example

For a telemedicine app, it extracts Must/Should/Could/Won’t requirements: Must—patient signup, video consults, prescriptions e-fax; Should—insurance verification; Could—asynchronous chat. It records NFRs (HIPAA, P99 latency < 300 ms, $2k/mo infra cap) and success metrics (first consult under 5 minutes).

Scenario

Outcome is a conflict-free scope that prevents scope creep. The subsequent SPEC links each design choice (e.g., selecting a regionalized object store and encrypted DB) to explicit compliance and latency requirements. Engineering gets a prioritized backlog and acceptance criteria that QA can test.

Example

For a marketplace MVP, it proposes: (1) Hexagonal service slicing (Users, Listings, Orders, Payments), (2) Postgres with tables: users(id, email, hashed_password, role), listings(id, seller_id, title, price_cents, status), orders(id, listing_id, buyer_id, state, total_cents), payment_intents(id, order_id, provider, status, amount_cents), (3) Event bus topics: order.created, payment.succeeded, refund.requested, (4) REST endpoints with OpenAPI, plus webhook signatures. Includes PlantUML component and sequence diagrams, indexing strategy (btree on foreign keys; partial index on listings(status='active')), and migration plan.

Scenario

Engineers can scaffold services, DB migrations, and API stubs immediately. Ops can provision managed Postgres + a message broker. Product can simulate flows via sequence diagrams (buyer checkout, refund). The design also specifies idempotency keys, retry/backoff, and exactly-once semantics via outbox pattern.

Example

Creates a 6-week MVP plan with sprint-level milestones, a DOR/DOD checklist, and risk register: • Risks—3rd-party auth rate limits, PCI exposure, data residency. • Mitigations—circuit breakers, token caching, use of hosted payment provider, geo-sharded storage. • Operational runbooks—SLOs (availability 99.9%), SLIs (error rate, p95 latency), alerts, and incident severity ladder.

Scenario

Leadership gets predictable delivery (demoable scope each sprint). Compliance sees mapped controls (logging, encryption, key rotation). Support gets triage playbooks. The team can ship the MVP, measure it, and iterate without re-architecting.