Patents
Protecting the anomaly detection, secure AI, and procurement innovations behind our platforms.
AlphaSix Patent Portfolio
One issued U.S. patent and six pending applications. The five pending non-provisionals each began as a provisional filed 1 August 2025 and were converted to non-provisional applications on 3 August 2026; the non-provisionals claim priority to those provisionals. A further provisional application is pending and has not yet been converted.
How the portfolio maps to AlphaSix products
3 productsQuantumDrive aligns directly with the prompt-anonymization and self-tool-generation filings and aligns with the Elastic natural-language-interface filing through its connected-systems architecture. QuantumDrive can also integrate separately licensed Qato anomaly detections, but that integration does not transfer Qato’s patent coverage to QuantumDrive. Quantify aligns directly with the two procurement filings and the conversational document-generation filing; it can additionally use prompt anonymization when the QuantumDrive privacy boundary is enabled. Qato is the direct commercial embodiment of U.S. Patent No. 9,866,578.
| Application No. | Subject | QuantumDrive | Quantify | Qato |
|---|---|---|---|---|
| 9,866,578 | Anomaly risk scoring | ○ | – | ● |
| 19/763,435 | Natural language interface, dashboards | ○ | ● | – |
| 19/763,469 | Prompt data anonymization | ● | ○ | – |
| 19/763,418 | Real-time self-tool generation | ● | – | – |
| 19/763,440 | RFI / RFP real-time feedback | – | ● | – |
| 19/763,401 | RFP value scoring | – | ● | – |
| Provisional pending | Conversational document generation | – | ● | – |
| Direct product alignments | 2 | 4 | 1 | |
| Conditional or integration | 2 | 1 | 0 |
● direct product alignment ○ conditional or integration relationship – not currently relied on
Issued patent
1 filingSystem and Method for Network Intrusion Detection Anomaly Risk Scoring
Establishes a behavioral baseline for each server across multiple independent parameters — "facets" — then measures live usage against those baselines to produce an abnormality value per facet and a combined anomaly risk score. The score tells administrators which nodes are most likely compromised or misconfigured, so analysts can focus attention inside data volumes too large to review by hand. This is the patent behind the Qato anomaly detection engine, with applications beyond cybersecurity in fraud detection, pharmaceutical diversion, and financial services.
Pending — AI platform & data infrastructure
3 filingsThree applications covering how non-technical users reach complex data through an LLM, and how sensitive data is kept out of the model while they do it. AS-002 and AS-005 are complementary: one turns plain language into dashboards over an existing cluster, the other lets the model build its own connector to a data source it has never seen.
Systems and Methods for Natural Language Interface for Elastic Cluster Management and Dashboard Automation
A user types a plain-language question; an LLM parses it for intent and entities, generates the corresponding domain-specific query, and runs it against an Elasticsearch cluster. The system then reshapes the returned data, has the LLM choose appropriate visualization types, and deploys a populated dashboard — no query language or Kibana configuration required. Conversation memory lets users refine the dashboard or add metrics through follow-up prompts.
Systems and Methods for Anonymizing and De-Anonymizing Sensitive Data in Prompts for Large Language Models
Before a prompt leaves the trusted environment, the client scans it for sensitive values — credentials, API keys, infrastructure identifiers, PII, regulated records — encrypts each one deterministically, and substitutes a placeholder. The third-party model never sees the real values, but because identical inputs always produce identical ciphertext, it still recognizes when the same entity recurs and reasons over the relationships correctly. The model's response comes back carrying the placeholders, and the client decrypts them locally to restore the original data.
Systems and Methods for Real-Time Self-Tool Generation for Connecting a Large Language Model (LLM) to a Data Source
Instead of relying on a predefined toolset, the LLM writes its own. It interprets a natural-language request to identify the target data source and the operations needed, drafts a tool specification, generates the executable code — a REST wrapper, a database connector — registers it in a tool registry, and runs it in an execution environment to return results. This removes the software-development cycle normally standing between a user and an unfamiliar data platform, whether that is an Elastic cluster, a SQL database, a GraphQL endpoint, or any API-driven system.
Pending — Procurement & proposal evaluation
2 filingsTwo applications in the same problem space, filed as separate families. AS-001 claims the real-time feedback loop that lets a vendor improve a submission before the deadline; AS-004 claims the evaluation architecture on the buyer's side — faceted parsing, retrieval-augmented scoring, price reasonableness, and an audit trail.
Systems and Methods for Providing Real-Time Feedback and Scoring for Requests for Information and Requests for Proposal
The buying entity supplies its evaluation criteria, which configure an LLM. As vendors enter their RFI or RFP responses, the model analyzes the submissions against those criteria in real time and returns feedback — dynamic scores, rankings, and written analysis — to the submitting vendor, who can then revise before the process closes. At the close, the system issues final scoring and rankings to the entity for its selection decision, making the basis of each score visible to both sides.
Systems and Methods for Requests for Proposal Value Scoring
A data processing module breaks each proposal into bounded independent facets that fit within an LLM's context window, then distributes them across GPU nodes for parallel inference — avoiding both truncation and attention dilution on long documents. A retrieval-augmented generation pipeline pulls context passages from a vector store into each prompt to constrain hallucination, and the inference engine extracts scoring insights facet by facet, including an assessment of whether the proposed price is reasonable. A storage and audit module logs every result and every piece of feedback, which is what makes the output defensible in government procurement.
Pending — Conversational document generation
1 filingOne further provisional application is pending in conversational document generation. Its title, application number, and description will be added to this page when the application publishes.