Arithmiks
Product · Contract IntelligenceClauseLens

Contract review you can check line by line

AI contract review that flags risks and extracts key terms in about a minute, with every finding traced to its source sentence.

Visit ClauseLens Explore features
ClauseLens review screen: a contract page with flagged clauses beside risk findings
  • Open AI
  • Anthropic
  • Gemini
  • Groq
  • Docling
  • Qdrant
  • Python
  • FastAPI
  • Celery
  • Redis
  • LibreOffice
  • Next.js
  • Tailwind CSS
  • PostgreSQL
  • MinIO
  • Docker Compose
  • nginx
  • FastEmbed
Overview

What is ClauseLens?

ClauseLens is an AI contract review and document intelligence platform for teams who read agreements for a living. Upload a contract and it returns a structured, source-grounded analysis: document type, risky clauses, commercial terms, a plain-English summary, and answers to follow-up questions.

The gap it closes

One commercial contract takes a trained reviewer two to four hours, most of it spent finding standard clauses and checking them against market norms. Clause libraries miss anything phrased differently, and generic AI chat answers without showing its source. ClauseLens returns a first-pass review in about a minute, with every finding linked to the sentence behind it.

ClauseLens contract library: analysed contracts as cards with risk counts
Why you can rely on it

If there is no source, there is no claim

In legal work a wrong answer is worse than no answer, so every finding ClauseLens shows can be traced to the sentence it came from.

Grounded, or dropped

Every extracted clause, term and obligation is re-matched against the real document text. If the source can't be found, the claim never reaches the reviewer.

Rules back the model

An LLM scores each clause against a configurable taxonomy, and a deterministic rules engine overrides it when a hard rule matches, like a missing liability cap.

Tested before it ships

Every model and prompt change runs against a golden dataset with an LLM-as-judge harness and an A/B runner, so regressions are caught before a real contract.

~1 min
from upload to a fully checked report
~4¢
AI cost per document
97%
retrieval accuracy
0
valid findings dropped after the grounding fix
The flow

What happens behind every review

Each contract is parsed, searched and analysed by specialized AI stages, and every finding is checked against the real text before a reviewer sees it.

Review a contractUpload contractsPDFDOCXBatchOne file or a whole batch, straight into your library
ParseDocument
Every file becomes a PDFWord files converted by headless LibreOffice, so one pipeline reads everything
Layout rebuiltSections, headings and tables restored; OCR only on low-confidence pages
Clause-aware chunks§12 → 12.3 → 12.3(a), each tied to its exact place on the page
Built onDoclingLibreOffice
AnalyseAI agents
Definitions agentDefined terms resolved so every later stage reads them the same way
Extraction agentClauses, key terms, obligations and a summary pulled from the text
Risk agentEach clause rated red, amber or green, with a rationaleRules engine can override
Model routerOpenAIAnthropicGeminiGroq
Source
found?Every citation re-matched to the real textNoFinding dropped
No
Finding dropped
ReportReview
FindingsRated risks, each linked to the clause and page it came from
Key terms and obligationsParties, dates, amounts and duties in tables you can check
SummaryA plain-language overview of the whole contract
Review and add notesReviewed contract
Ask your contractsAsk a questionOne contractWhole libraryPlain English, follow-ups welcome
SearchRAG
Hybrid retrievalSemantic and keyword search, fused and reranked before the model sees anything
IndexQdrantFastEmbed
AnswerAI
Evidence onlyFollow-ups rewritten against the conversation; the model answers from retrieved text, not memory
CiteGrounded
Answer with citationsEvery statement links to its page and clause. The same grounding check applies, so an unsupported claim is dropped, not shown.
Highlights

Key features

1. Grounded Risk Findings:

Every clause rated against market norms and traced to its source sentence.

ClauseLens findings: a flagged clause with why it was flagged, market norm and suggested position
More from Arithmiks

Keep exploring

View all case studies
Query your database in plain English conversationsQuantaBI PlatformQuery your database in plain English conversationsA multi-tenant BI platform that lets any team query their own database in plain English, no SQL required.Read case study AI-Powered Marketing Automation PlatformGOMachine LearningAI-Powered Marketing Automation PlatformAn AI-powered platform that automates digital marketing—handling content, publishing, and ads to help agencies grow.Read case study

This project drew on our AI & Data Solutions service.

Describe your idea

Tell us what you're building. We'll read it before the call, so the 30 minutes go on your problem rather than introductions.

My role (optional)
Optional · PDF, DOC, DOCX or PPT, up to 10 MB
Recognition
Upwork Top Rated PlusUpworkRank — #146 in Web Development, top 0.87%
Office
Lahore, Pakistan
Omer and his team put in significant effort and delivered many positive contributions.
Louis-AntoineGoAgents Founder & CEO