Legal Tech Startup Compresses 5-Day Patent Analysis Workflow Into 5 Minutes with Claude on Amazon Bedrock
At a glance
With 1.85 million annual patent searches globally, professional firms require 2–5 days of manual analysis per assessment at $700+, putting thorough patent research out of reach for individual inventors and small businesses.
New Math Data compressed NLPatent’s 12-month internal launch timeline to 13 weeks, delivering a production-ready agentic AI system designed for 100M+ TPM that delivers attorney-quality patentability analysis that cuts 2–5 days of manual review to ~5 minutes at a fraction of traditional search cost.
The system launched in May 2026 and is designed for 8 planned workflow expansions across NLPatent’s IP platform.
Industry
Use Case
- Natural Language Processing
- Agentic Workflow & Process Automation
- Legal & Contract Management
Solution implemented
- Claude Sonnet on Amazon Bedrock for legal reasoning generation, selected for long-context handling and accuracy on element-level patent claim comparison; other Claude models under evaluation for production cost optimization
- Claim-element relevance scoring: NLPatent's prior-art search-trained embedding model finds the closest matching sections per claim element, ranks by cosine similarity, and sends only the strongest prior-art candidates into deeper Bedrock reasoning
- Multi-stage agentic pipeline: patent retrieval, chunking, semantic scoring, NLI enrichment, and LLM reasoning orchestrated end-to-end via AWS Step Functions and Lambda
- Serverless parallel analysis: hundreds of prior-art references processed simultaneously via Lambda, workflow artifacts stored in S3, element thresholding and 98% Bedrock prompt-cache reuse keep costs viable at high scale
- VPC-peered architecture with vector search integration to NLPatent's patent database for low-latency retrieval at production scale
- Architecture designed for 100M+ TPM peak throughput, sized for projected global search volume across NLPatent's 8-workflow IP platform; internal testing reached ~60M TPM
- AWS granted a dedicated Bedrock quota increase to 80M TPM (40M Global + 40M EU), provisioned specifically to support NLPatent's production workload and growth trajectory
The value equation
- 6+ months faster to production: May 2026 vs. Q2 2027 internal timeline
- ~5 days to ~5 minutes workflow speed (99% reduction)
- $700+ → below competitor cost
- Attorney-quality yes/no decisions with element-level legal reasoning and cited prior-art references
- Architecture designed for 100M+ TPM peak throughput; internal testing reached ~60M TPM; AWS provisioned 80M TPM dedicated Bedrock quota to match production ambition
- Modular design enabling 8 planned workflow expansions across NLPatent's IP platform
Company Snapshot
NLPatent, a patent search and analysis SaaS platform enabling inventors and legal professionals to evaluate patent landscapes and assess patentability through AI-powered tools.
Location
Canada, United Kingdom
Customer Situation
Traditional patentability searches cost $700+ and require 2–5 days of manual attorney review, a bottleneck that limits access to the 1.85M annual patent search market. Attorneys manually review hundreds of patents, analyze claim-by-claim comparisons against prior art, and draft detailed legal reasoning for each element. The process is time-intensive and cost-prohibitive for individual inventors and small businesses, who represent a significant and underserved share of global patent filings.
NLPatent identified the automation opportunity but faced a 12-month internal timeline to build and launch their AI-powered patentability researcher, their top strategic priority for 2026. Every month of delay meant delayed revenue, competitive exposure, and a shrinking window to establish market position.
NMD Solution
New Math Data compressed Legalicity’s 12-month timeline to 13 weeks through AWS-funded team augmentation, delivering 6+ months faster time to production.
6-Week POC (Nov–Dec 2025): Validated technical feasibility, cost viability, and production architecture. Proved the core workflow compression from 2–5 days of manual analysis to ~5 minutes, and confirmed a viable cost per search using Claude Sonnet on Bedrock with 98% prompt-cache reuse.
7-Week Production Build (Jan–Mar 2026): Scaled POC to multi-tenant SaaS infrastructure with observability, compliance monitoring, token-cost controls, and modular architecture for 8 planned workflow expansions. Integrated NLPatent’s prior-art search-trained embedding model for claim-element relevance scoring, enabling the system to pre-filter prior art candidates before routing only the strongest matches into Bedrock reasoning, keeping latency low and costs viable at scale.
The architecture was designed from the outset for 300M+ TPM peak throughput, sized against NLPatent’s projected global search volume across their full IP platform. Internal testing reached ~60M TPM. AWS recognized the scale of the use case and provisioned a dedicated Bedrock quota increase to 80M TPM (40M Global + 40M EU), removing the throughput ceiling that would otherwise have constrained production launch.
What We Delivered
May 2026 production launch, 6+ months faster than Q2 2026 internal timeline.
The system delivers attorney-quality yes/no patentability decisions with element-level legal reasoning and cited prior-art references, analyzing hundreds of patents per workflow in ~5 minutes. The serverless architecture processes prior-art references in parallel via Lambda, stores reproducible workflow artifacts in S3, and uses element thresholding plus 98% Bedrock prompt-cache reuse to keep costs viable as volume scales.
The quota increase secured by AWS is as much a validation of the architecture as it is an infrastructure decision. NLPatent enters production with headroom to scale, a modular foundation supporting 8 planned workflow expansions, and an AWS partnership that has already demonstrated its commitment to the use case.
Ready to Transform Your SaaS Business?
See how New Math Data can transform your SaaS business with AWS-powered innovation.