Proforma Intelligence

What It Is

A web-based financial diagnostic that runs company intake through a 5-step analysis pipeline and streams a synthesis report. Steps 2-3 are deterministic TypeScript with zero AI calls -- signal scoring against validated thresholds, benchmark comparison against industry databases. Steps 4-5 use AI Gateway with Claude models for hypothesis generation and narrative synthesis. Built on Next.js 16, React 19, SQLite, AI SDK v6, and shadcn/ui. Replaces a manual CustomGPT v5 workflow.

Why It Matters

Rigorous financial analysis requires deterministic computation to constrain what AI can say. Most AI-powered financial tools are fully generative -- the model produces scores, benchmarks, and recommendations with no ground truth to check against.

Proforma Intelligence splits the pipeline. Signal scoring thresholds come from validated knowledge bases stored as JSON, not from model inference. Benchmark comparisons come from curated databases (benchmarks.json), not from generated estimates. Steps 2-3 produce deterministic, auditable, reproducible results with zero AI calls. AI enters only at Steps 4-5, where it generates hypotheses and narrative synthesis within constraints established by the deterministic steps. The model cannot contradict the signal scores because it receives them as input, not as a generation target. This is Constraint Surface Governance applied to financial analysis: AI operates within a boundary defined by validated data, not in open space.

Proof Points

5-step pipeline: intake then signal scoring then benchmark comparison then AI synthesis then report streaming
Steps 2-3: zero AI calls -- deterministic TypeScript computation against signal-rules.json and benchmarks.json
Steps 4-5: Sonnet for classification, Opus for synthesis via Vercel AI Gateway
Knowledge base: validated signal rules, benchmarks, anonymized client examples -- all version-controlled JSON
Built on Next.js 16, React 19, SQLite (better-sqlite3), AI SDK v6, shadcn/ui, Vitest
Analytical voice contract imported verbatim from system_instructions.md -- output tone matches the existing CustomGPT v5 standard
Streamed report generation with real-time progress indicators
Replaces a manual workflow that took days with a diagnostic that runs in minutes

Market Position and IP

Patent-protected methodology (USPTO 19/418,922). The hybrid deterministic-AI architecture is unique in the financial diagnostic market. Competing tools (ChatGPT financial analysis, Copilot business intelligence, custom GPTs) are fully generative -- no deterministic constraints on what they produce. Proforma Intelligence guarantees that signal scores and benchmarks are computed from validated sources, not inferred. The signal scoring thresholds and benchmark databases represent years of consulting engagement data, creating a proprietary data moat. The architecture pattern (deterministic steps constraining AI steps) is replicable across any domain requiring auditable AI output.

Novel Research Contribution

Existing research on AI-assisted financial analysis treats the entire pipeline as a generation problem. Proforma Intelligence demonstrates the constructive proof that hybrid deterministic-AI pipelines produce auditable results where fully generative pipelines cannot. The 5-step architecture provides a replicable pattern for any domain where AI must operate within validated constraints. The analytical voice contract shows that output quality can be specified as a formal input, not hoped for as an emergent property.

Implementation and Impact

Clients receive a financial diagnostic delivered through the Proforma Intelligence web app. The streamed report includes signal scores with threshold explanations sourced from validated knowledge bases, benchmark comparisons with industry context from curated databases, root cause hypotheses with confidence ratings, and prioritized initiative recommendations. For PE operating partners, this replaces weeks of manual analysis with a diagnostic that runs in minutes and produces results that can be audited and defended. Measurable outcome: financial diagnostic with deterministic signal scoring, auditable benchmarks, and AI-constrained synthesis -- not AI-generated numbers.

Connections

Imperatives: Constraint Surface Governance
Papers: cost-of-rationality
Frameworks: Proforma Engine, Agentic OpModel
Capabilities: Financial Value Creation and Capture