AI requirements writer

Shall-statements with rationale, verification, and receipts.

Generic LLMs produce shall-statements that read fine and don’t hold up under review. Blueprint produces shall-statements that read fine and are traceable to the source brief, the parent NGO, and the NASA SE Handbook §4.2 section that justified the structure. Verification method per shall. Rationale per shall. Citation per claim.

Start free →See the full pack

What a Blueprint shall looks like.

REQ-014: The vehicle shall maintain a dry mass of less
than or equal to 8 kg.

  Type:           Performance
  Verification:   Analysis (Handbook §4.2.4)
  Level:          system
  Rationale:      Launch-vehicle 12 kg upper-stage allocation
                  with 4 kg propellant margin per source brief.
  Traces to:      NGO-3 (mass budget)
  Cited from:     [DC-7] "Upper stage allocation of 12 kg dry
                  mass leaves 4 kg propellant margin..."

Every field is grounded. Type comes from the four NASA categories (Functional, Performance, Interface, Constraint). Verification method is one of T / A / I / D per Handbook Table 4.2-2. Level is back-filled from the Allocated Baseline. The rationale is extracted from the cited DocChunk; the chunk text is visible in the artifact viewer.

The math.

80%

citation precision

Independent LLM judge over each Requirement → DocChunk link. Measured on the ARCHIPELAGO smoke.

T / A / I / D

verification methods

Test / Analysis / Inspection / Demonstration. Per shall, per the NASA Handbook canon.

agent roles

Requirements Worker runs against the Stakeholder Worker's NGO output, with the Citation Verifier judging both.

$0.17

per Phase 1.4 run

LLM cost, measured on DeepSeek backbone. Generation is rounding error against your billable rate.

Versus the alternatives.

	Hand-writing	Generic LLM prompt	Blueprint
Wall time	80 hours	2 hours	15 minutes
Source citations	By discipline	No	80% measured
Verification method per shall	By discipline	Inconsistent	Per Handbook §4.2
Traces to parent NGO	By discipline	No	Graph-RAG
Rationale per shall	By discipline	Plausible-sounding	Cited
Phase-D readiness	No	No	Yes

“Generic LLM” here is GPT-4o or Claude given the brief + a freeform prompt. The output reads fine; it does not survive a structured review.

Questions about requirements generation.

How many shall-statements does Blueprint produce per brief?

Scales with the brief. ARCHIPELAGO (1,500-word coastal-swarm brief) yields ~25 functional + ~12 performance + ~8 interface + ~6 constraint requirements. Larger briefs produce more; the structure stays NASA-canonical.

Can Blueprint enforce my organization's house style for shall-statements?

The output is markdown — open the requirements doc in your editor and apply house style. Future: a per-project style template is roadmapped, not yet shipped. For now, the post-generation edit is the route.

What about derived requirements (DTRs)?

Blueprint's Decomposition Worker emits DTRs as a separate artifact, with each DTR tracing back to the parent requirement and forward to the allocated component. Same citation discipline as the parent shall-statements.

Related.

Automated ConOps generation from a mission brief →
Operational scenarios with nominal + off-nominal paths, generated against Handbook §4.1.
AI shall-statement writer (deep dive) →
The shall-statement structure, what the verification column means, and how to read a Blueprint citation.
← Back to: Blueprint for solo SE consultants
The full pack — Phase 1.1 through Phase D readiness.

Try it on a brief you already have.

Free tier · no card · 3 runs / month. Run one against an old engagement’s brief and inspect the citation chain.

Start free →