ClauseCraft AI — Contract Review API

Concept

ClauseCraft AI is a B2B SaaS backend that gives legal teams a programmable, AI-powered contract review pipeline. Upload a PDF or DOCX, and within seconds the API returns an executive summary, ranked risk items with verbatim clause excerpts, and AI-generated redraft suggestions — all benchmarked against the organisation's own "playbook" of pre-approved standard clauses.

Target audience: In-house legal teams at SMEs, startup legal ops, boutique law firms.

Value proposition: Converts a multi-hour manual review into a sub-minute API call, freeing lawyers for strategic work.

---

Architecture

Client (browser / legal tool)
        │  JWT Bearer
        ▼
┌─────────────────────────────────────────────────────┐
│  Spring Boot 4 / Java 25                            │
│                                                     │
│  AuthController   ContractController  PlaybookCtrl  │
│        │                │                  │        │
│  UserService    ContractService    PlaybookService  │
│        │                │                  │        │
│  UserRepository  ContractRepo  PlaybookClauseRepo   │
│                       │                            │
│           DocumentParserService                     │
│           (DOCX/PDF/TXT text extraction)            │
│                       │                            │
│             AnalysisService ◄── PlaybookService     │
│                       │  (RAG context assembly)    │
│                       ▼                            │
│              ┌──────────────────┐                  │
│              │  LlmClient       │  OpenFeign        │
│              │  (Feign + JWT)   │────► OpenAI API   │
│              └──────────────────┘                  │
│                       │                            │
│           AnalysisResultRepository                  │
│                                                     │
└─────────────────────────────────────────────────────┘
        │
  PostgreSQL (JPA / Hibernate 6)

Key design decisions

Concern	Choice	Reason
Auth	JWT (jjwt 0.12.6)	Stateless, scales horizontally
LLM integration	OpenFeign → OpenAI API	Swappable to any OpenAI-compatible endpoint
Document parsing	Pure JDK (ZIP + regex)	Zero extra dependencies; production should add Apache PDFBox
Analysis persistence	JSON TEXT column	Schema-free storage of evolving LLM responses
Concurrency	Virtual threads (Java 25)	I/O-heavy LLM calls benefit from lightweight threads
Playbook context (RAG)	In-DB retrieval → prompt injection	Simple RAG without a vector DB; scales to PGVector later

---

Endpoints

All endpoints require %%INLINE0%% except %%INLINE1%%.

Authentication

Method	Path	Body	Response
%%INLINE2%%	%%INLINE3%%	%%INLINE4%%	%%INLINE5%% (201)
%%INLINE6%%	%%INLINE7%%	%%INLINE8%%	%%INLINE9%% (200)

Contracts

Method	Path	Description
%%INLINE10%%	%%INLINE11%%	Upload document (%%INLINE12%%, field %%INLINE13%%)
%%INLINE14%%	%%INLINE15%%	List user's contracts
%%INLINE16%%	%%INLINE17%%	Contract metadata
%%INLINE18%%	%%INLINE19%%	Trigger AI analysis (calls LLM)
%%INLINE20%%	%%INLINE21%%	Retrieve analysis report

Analysis response shape

{
  "contractId": 42,
  "fileName": "nda_acme.pdf",
  "analyzedAt": "2026-05-23T14:32:00",
  "executiveSummary": "...",
  "overallRiskLevel": "HIGH",
  "keyObligations": ["Party A shall...", "..."],
  "importantDates": [
    {"description": "Contract expiry", "date": "2027-01-01"}
  ],
  "risks": [
    {
      "clauseTitle": "Limitation of Liability",
      "clauseText": "...",
      "riskLevel": "HIGH",
      "description": "Cap is capped at $500 which is below contract value.",
      "suggestedRedraft": "Limit liability to no less than fees paid in the 12 months preceding the claim."
    }
  ],
  "playbookCompliance": {
    "compliant": false,
    "missingClauses": ["FORCE_MAJEURE"],
    "nonCompliantClauses": ["TERMINATION"]
  }
}

Playbook

Method	Path	Description
%%INLINE22%%	%%INLINE23%%	List user's clause templates
%%INLINE24%%	%%INLINE25%%	Add a pre-approved clause
%%INLINE26%%	%%INLINE27%%	Single clause
%%INLINE28%%	%%INLINE29%%	Update clause
%%INLINE30%%	%%INLINE31%%	Remove clause

Playbook clause types

%%INLINE32%% · %%INLINE33%% · %%INLINE34%% · %%INLINE35%% · %%INLINE36%% · %%INLINE37%% · %%INLINE38%% · %%INLINE39%% · %%INLINE40%% · %%INLINE41%% · %%INLINE42%% · %%INLINE43%%

---

MVP Features Coverage

Feature	Implementation
Contract upload + text extraction (PDF, DOCX)	DocumentParserService — ZIP/XML for DOCX, regex scan for PDF
AI risk analysis — flags non-standard/ambiguous clauses	%%INLINE45%% → %%INLINE46%% with structured prompt
Playbook comparison (RAG step)	PlaybookService.buildPlaybookContext() injected into LLM prompt
Executive summary generation	LLM returns %%INLINE48%% + %%INLINE49%% + importantDates
Redraft suggestions per risky clause	LLM returns %%INLINE51%% per %%INLINE52%% item

---

Running Locally

Prerequisites

• Java 25+
• PostgreSQL 16+ (or Docker)
• Maven 3.9+

1. Start PostgreSQL

docker run -d \
  --name clausecraft-db \
  -e POSTGRES_DB=clausecraft \
  -e POSTGRES_USER=clausecraft \
  -e POSTGRES_PASSWORD=clausecraft \
  -p 5432:5432 \
  postgres:16

2. Configure application.yml (add datasource + clausecraft keys)

spring:
  datasource:
    url: jdbc:postgresql://localhost:5432/clausecraft
    username: clausecraft
    password: clausecraft
  jpa:
    hibernate:
      ddl-auto: create-drop   # use 'update' in production
    show-sql: false

clausecraft:
  jwt-secret: "YOUR_BASE64_SECRET_MIN_32_BYTES"
  jwt-expiration: 86400000
  llm:
    base-url: "https://api.openai.com/v1"
    api-key: "sk-..."
    model: "gpt-4o"

No API key? Leave api-key blank. The analysis pipeline returns a stub response so you can test the full upload/retrieve flow without cost.

3. Build & Run

cd solutions/2026-05-23-clausecraft-ai-api
mvn clean package -DskipTests
java -jar target/clausecraft-ai-api-0.0.1-SNAPSHOT.jar

4. Quick test

# Register
curl -X POST http://localhost:8080/api/v1/auth/register \
  -H "Content-Type: application/json" \
  -d '{"name":"Alice","email":"alice@example.com","password":"password123"}'

# Upload contract
TOKEN="<jwt from register>"
curl -X POST http://localhost:8080/api/v1/contracts \
  -H "Authorization: Bearer $TOKEN" \
  -F "file=@/path/to/contract.pdf"

# Analyse (returns full report)
CONTRACT_ID=1
curl -X POST http://localhost:8080/api/v1/contracts/$CONTRACT_ID/analyze \
  -H "Authorization: Bearer $TOKEN"

---

Business Analysis

Revenue model

Tier	Limit	Target
Starter	10 reviews/month	Solo lawyers, bootstrapped startups
Professional	Unlimited reviews + team	Mid-size legal teams
Enterprise	Custom playbook automation + analytics	Corporates, law firms

Moat

1. Playbook personalisation — each company's clause library makes the AI progressively smarter for that client; switching cost is high.
2. Audit trail — every analysis is persisted with timestamps; critical for compliance evidence.
3. Vertical depth — the prompt engineering and risk taxonomy are domain-specific; generic AI wrappers can't match.

Growth path

• Phase 1 (MVP): REST API + web UI for contract upload and report viewing.
• Phase 2: PGVector integration for semantic clause similarity search (true RAG).
• Phase 3: Clause negotiation workflow — track redline history against playbook versions.
• Phase 4: Integrations — Salesforce, DocuSign, Slack alerts on HIGH-risk clauses.

---

References

• Best AI SaaS Product Ideas 2026 — Groovyweb
• Spring Boot 4.0 Reference
• Spring Security 7 — JWT Configuration
• OpenAI Chat Completions API
• jjwt 0.12.x Migration Guide

---

Análisis FinOps para ClauseCraft AI

Visión General

ClauseCraft AI, una micro-startup en el nicho de Legal-Tech, demuestra un modelo de negocio con un potencial de rentabilidad excepcionalmente alto en sus etapas iniciales. La eficiencia inherente de los servicios de IA, combinada con una infraestructura cloud lean, permite mantener los costos operativos bajos mientras se genera valor significativo para los clientes.

Desglose de Costos Mensuales Estimados

Para una operación inicial con un volumen de ~5 millones de tokens LLM al mes y una base de clientes en crecimiento, los costos se distribuyen de la siguiente manera:

• Costos de LLM (OpenAI gpt-4o): $30/mes

* Estimación basada en el procesamiento de aproximadamente 500 contratos al mes (con un promedio de 10,000 tokens por análisis, incluyendo prompt, playbook y respuesta). Se asume una división de 4.5M tokens de entrada y 0.5M tokens de salida utilizando gpt-4o ($0.005/1K input, $0.015/1K output).

• Infraestructura Cloud (AWS Baseline): $55/mes

* EC2: %%INLINE56%% (ej. %%INLINE57%% para la aplicación Spring Boot). * RDS: %%INLINE58%% (ej. %%INLINE59%% PostgreSQL para la base de datos de contratos y playbooks). * S3: $2/mes (almacenamiento de documentos, ~100GB). * Networking/Load Balancer/VPC: $10/mes (ALB básico para HTTPS y tráfico). * Monitoring (CloudWatch): $5/mes (logs y métricas esenciales).

• Overheads SaaS y Operativos: $120/mes

* Procesamiento de Pagos: $50/mes (ej. tarifas de Stripe para $1750 de ingresos). * Herramientas de Marketing/Ventas Básicas: $50/mes. * Dominio, Email, Misceláneos: $20/mes.

Costo Operativo Total Estimado: $205/mes

Ingresos y Margen de Beneficio Estimados

Basándonos en el modelo de monetización de suscripción por niveles y una base de clientes inicial para una micro-startup:

• Clientes Starter: 10 clientes @ $75/mes = $750/mes
• Clientes Professional: 5 clientes @ $200/mes = $1000/mes

Ingreso Mensual Total Estimado: $1750/mes

Beneficio Mensual Neto: $1750 (Ingresos) - $205 (Costos) = $1545

Margen de Beneficio (Profit Margin): (($1750 - $205) / $1750) * 100 = 88%

Conclusión FinOps

ClauseCraft AI muestra un margen de beneficio operativo excepcional del 88% en su fase de micro-startup. Este alto margen es una fortaleza clave, impulsada por la eficiencia de los servicios de IA y una infraestructura cloud optimizada. A medida que la startup escale, será crucial monitorear la relación costo-beneficio de los LLM y la infraestructura, especialmente si se migra a modelos más costosos o si el volumen de tokens aumenta drásticamente. La implementación proactiva de estrategias FinOps, como las optimizaciones mencionadas, garantizará que ClauseCraft AI mantenga su alta rentabilidad y escalabilidad a largo plazo.