About Project
Amunet IP is an end-to-end platform for patent research and analysis created for attorneys, analysts, in-house teams, and even solo inventors who have to work with intellectual property every day. The platform gives them a faster, smarter way to handle patents.
Traditionally, the process looked like this: find 1,000+ patents manually, read through them, highlight overlaps, score them, build charts. All by hand. All painfully slow.
Now? Amunet IP does the same in under 15 minutes just by adding an AI assistant to its core. Users can search, compare, visualize, and evaluate patents without having to go through each of them manually.
We teamed up with Amunet to bring this idea to life through a full suite of web development services. Starting from scratch, we built everything from the UX and backend infrastructure to the NLP and AI query systems, optimizing the flow of massive patent datasets for real-time insight. The result is lean, fast, and built to scale without burning through compute or confusing the user.
Interesting Facts:
Amunet’s users include patent attorneys, analysts, IP brokers, solo firms, and valuation experts, basically anyone who lives inside claim charts and patent PDFs.
They’re based in Florida — yeah, the land of gators and wild headlines — but also quietly turning into a real tech and innovation hotspot in the Southeast.
The platform’s modular design bypasses AI context window limits. That’s how it handles up to 1,000 patents per query without breaking a sweat, thanks to a smart microservices setup.
With optimized NLP filtering, we cut LLM costs from ~$200 to just ~$20 per run.
Client Request
The Amunet team came to Lumitech with a big vision: make patent analysis smarter, faster, and actually usable for busy professionals. They weren’t looking for a flashy AI demo or a bloated enterprise tool. They needed a real working product that could handle huge IP workloads without falling apart or racking up a $10K cloud bill.
The core challenge was scale. A single query could involve thousands of patents, each packed with dense technical language. Traditional tools couldn’t keep up, and GPT-powered solutions risked becoming too expensive or too slow. And the UI? It had to make sense to legal professionals, analysts, and IP-focused companies. Meaning anyone working under pressure, without time to learn a new complicated UI and UX design system.
So we rolled up our sleeves. Our goal was to build a platform that looked simple on the surface, but ran like a machine under the hood. We designed a modular architecture, used NLP models to pre-filter and score content before AI ever got involved. Then, we coordinated multiple AI agents to handle data retrieval, analysis, and final output.
The result is a powerful, stable, and surprisingly lightweight system that’s now helping users get weeks' worth of patent insight in just minutes.
Amunet wanted something users could trust. We helped make that happen.
Approach
This wasn’t about showing off some cutting-edge AI. It was about building a platform people actually can use and, most importantly, want to use — because it works every time and doesn’t melt under pressure.
Here’s how we did it:
Dug deep into the patent research workflow to figure out what’s painful and what can be automated.
Worked closely with the Amunet team to define features that work across firms, roles, and use cases.
Designed a modular backend with microservices so big workloads wouldn’t jam up the system – crucial approach in development for the finance industry.
Used smart prompt engineering and NLP pipelines to reduce LLM calls to only the stuff that really matters.
Shipped fast, iterated often, and kept things lean the whole way through.
What came out of that? A platform that feels easy but thinks hard. And that’s exactly what Amunet users needed.
Challenges
Making big data manageable Comparing thousands of patents can crash most systems. We built Amunet with microservices that split and conquer. Each part of the system handles one piece of the job and passes it along. That keeps things fast, even under heavy use.
AI costs spiraling out of control LLMs like GPT are powerful, but also expensive. Especially if you send them giant patent files. So we didn’t. We used NLP to filter and pre-process everything, only passing essential chunks to the AI. End result: cost down, speed up.
The context window problem Most AIs can only “see” a limited amount of text at once. We solved that by using multiple AI agents: one to grab the data, one to break it down, one to write the results. Each doing their part, then handing it off. It's like a relay team, but for patent logic.
Designing for legal minds, not tech pros Lawyers don’t want to read API docs. So we built a UI that gives them results in plain language, with clear visuals and exportable reports. No complications. Just tools that do what you ask.
Core Features
Fast & Flexible Patent Analysis At the heart of Amunet is its ability to turn complex, high-volume patent queries into clear, structured insights in minutes. Users can input a request, say, “Compare recent filings in machine learning-based fraud detection” – and the system kicks off a full analysis pipeline: pulling relevant documents, ranking them by match strength, and generating a digestible report. What used to take two weeks now takes 10-15 minutes, with no analyst backlog, no manual scoring, and no email chains.
AI-Powered Cockpit Interface At the surface, Amunet offers a clean, minimal UI where users can type in natural-language queries like “Find patents relevant to autonomous drone navigation” or “Show potential overlaps with US12345678.” Behind the scenes, this input kicks off a sequence of AI operations, from data retrieval to claim scoring, but the user only sees what matters: results, visualizations, and suggested next steps.
Modular AI Agent Architecture Instead of overloading a single model with everything at once, we split the work into discrete agents. One handles SQL generation and database retrieval, another parses the patent text and breaks it into comparable structures, a third one evaluates and compiles insights into a digestible format. This modular structure avoids LLM context limits and makes the system far more stable under heavy use.
NLP-Driven Categorization & Pre-Filtering Before any data hits the LLM, we run it through NLP pipelines that sort, tag, and prioritize patents based on structure, domain, and keyword clustering. This cuts down the LLM payload, ensuring we only send what’s essential. For users, it means they don’t waste time on irrelevant patents. And for the platform it means drastically lower GPU costs.
AI Claim Chart Assistant Users can upload two or more patents or technical documents and ask the system to identify overlaps, structural differences, or conflicting language. The AI doesn’t just compare raw text, it interprets claim logic, aligns comparable elements, and presents results in editable, exportable formats. It’s like getting a junior patent associate who never sleeps and color-codes everything.
Visual Portfolio Mapping Once results are in, users can view them as interactive visualizations: clusters by technology domain, heatmaps of licensing potential, or overlap zones with competitors. These visuals aren’t static either: they respond to user prompts like “Highlight gaps in our drone navigation claims” or “Filter for energy tech with high citation density.”
Scalable Microservices Architecture To support heavy concurrent use (think: 1,000+ patent comparisons per query), we broke the system into microservices. Each service runs independently, handling a specific task. That means faster responses, better uptime, and zero crashes when someone runs a monster query at 3 a.m.
Summary
Amunet IP was one of those projects that just clicked. The idea was sharp, the challenge was real, and the solution needed to work at scale, with zero margin for fluff.
We helped build something that’s now changing the way people handle patents. No more spreadsheets. No more weeks of manual review. Just smart tools that help people move faster.
If you’re building in legal tech, AI, or anything that involves big decisions and even bigger data – we’re your team.
Let’s talk and make it happen.
