How to manage aircraft leases with AI agents

How to manage aircraft leases with AI agents

Component | Article_teaser

September 25, 2026

4

-minute read

Artificial intelligence

Michael Dawson

Bertrand Rault

Bala Sista

Indranil Paul

Subpattern | social_share_set

Component | social_share

Share

Subpattern | subscribe

Component | social_icon

Subscribe to RSS

Component | Icon

© Red Hat, Inc. CC-BY-4.0 licensed

Subpattern | results_nav

Group layout

Component | Nav_links

• Back to all posts

Component | Generic

AI quickstarts are a catalog of ready-to-run, industry-specific use cases for your Red Hat AI environment. Each aims to provide a simple and practical way to solve real-world problems using AI on enterprise open source infrastructure.

This quickstart takes on a document problem most people outside of aviation never see: the aircraft lease.

Aircraft leases are typically dense, complex, and hundreds of pages long. Reading and analyzing them is a job for a specialist, and it can take several days. NeIO LeasingOps, built by Codvo.ai , does a 1st pass with a set of AI agents, handing people structured results they can act on.

The problem: Lease contracts are hard to read at scale

Aircraft leases are some of the densest commercial documents around. One agreement can include:

• Variable payment schedules and financial terms.

• Maintenance reserve calculations tied to flight hours and cycles.

• Return-condition requirements with exact technical specifications.

• Obligations spread across lessors; lessees; and maintenance, repair, and overhaul (MRO) providers.

• Compliance requirements under multiple regulators and standards bodies like the U.S. Federal Aviation Administration (FAA), European Union Aviation Safety Agency (EASA), and the International Air Transport Association (IATA).

• Amendments that change the base terms over time.

A team of specialists will typically extract the terms, reconcile them against operational data, and validate everything by hand. It works, but it’s slow. It doesn’t scale, and it ties up expensive expertise on repetitive reading. Decisions often get delayed because of it.

The approach: 10 steps, 10 agents

NeIO LeasingOps by Codvo.ai runs each document through 10 agents, one after another, from upload to recommendation. The agents are built with LangGraph. A background worker pulls jobs from a Redis queue and advances each document through the chain, writing results to PostgreSQL as it goes. The whole thing runs on Red Hat OpenShift AI.

Here’s what each stage does:

1. The contract intake agent validates the upload and classifies it as an operating lease, finance lease, wet lease, dry lease, maintenance reserve claim (MRC), amendment, or other and routes it down the right path.

1. The term extractor agent pulls structured data out of unstructured contract text using named-entity recognition and targeted prompts like:

• Key dates (start, end, option periods).

• Financial figures (rent, deposits, reserves).

• Asset identifiers (tail number, manufacturer serial number, registration).

• Parties (lessor, lessee, guarantors).

• Special clauses.

1. The obligation mapper agent finds every obligation and files it under a category—maintenance, payment, return condition, insurance, reporting—with a responsible party and a due date. The result is 1 obligation registry for the contract.

1. The utilization reconciler agent compares contracted flight hours and cycles against actual MRO data and flags the gaps. This feeds the reserve calculations and the compliance checks.

1. The reserve calculator agent tracks maintenance reserve contributions, drawdowns, and balances and spots shortfalls. It reads the financial clauses with the model but does the arithmetic deterministically, so the numbers hold up.

1. The variance detector agent flags where the contract and reality diverge: payment variances, maintenance-schedule slippage, condition-report mismatches.

1. The return readiness analyzer agent compares the contractual return conditions against the aircraft’s current condition reports, estimates redelivery cost, and produces a punch list at the end of the lease.

1. The evidence pack agent assembles audit-ready documentation that ties every finding back to the specific clause it came from—the part that matters when there’s an audit or a dispute.

1. The decision support agent pulls everything together into a return, extend, or buyout analysis with risk-adjusted recommendations.

1. Escalation is the final stage and human-in-the-loop checkpoint. Anything that needs a person to decide—a material variance, a compliance flag, a deadline coming due—gets routed to experts with the full evidence trail attached, along with the alerts and reminders that go with it.

How it’s built

NeIO LeasingOps is a cloud-native application that runs on Red Hat OpenShift. The application consists of a Next.js 15 frontend for uploading documents and viewing results, backed by a FastAPI service that handles document ingestion, contract CRUD operations, and the REST API. A Python worker processes documents through the agent pipeline off a Redis job queue, while a PostgreSQL database stores document records, extracted terms, and agent results. Redis manages the job queues and pipeline status.

For AI serving, the system uses vLLM to serve IBM Granite 3.3 2B, an Apache 2.0 open model from IBM that requires no Hugging Face token, with LlamaStack acting as an OpenAI-compatible gateway in front of vLLM. Document parsing is handled by Docling when a Docling service is available, falling back to PyMuPDF when needed. The entire system runs on either a central processing unit (CPU) for cost efficiency or a graphics processing unit (GPU) for improved performance, using the same configuration in both cases. vLLM, LlamaStack, and the Granite model come from the Red Hat AI architecture charts and deploy separately from the application.

The application, along with its PostgreSQL and Redis dependencies, ships as a Helm chart that deploys everything into a single namespace. You can use the in-cluster databases provided by the chart or point it at your own external instances. The system supports both CPU and GPU inference, which can be easily switched using configuration values.

What you get out of it

The system delivers significant speed improvements, reducing analysis tasks that previously took analysts days down to minutes on a GPU, or slightly longer on CPU-only configurations. It scales without interruption from validating a single document to working through an entire portfolio.

The pipeline supports consistency by applying the same analysis to every document, eliminating the run-to-run variability inherent in manual review. Multipass extraction with schema validation keeps the data clean and reliable throughout the process.

Every extracted fact comes with an audit trail through the evidence pack agent, which links each finding back to its source clause. This provides the documentation needed if regulators or counterparties request justification for specific conclusions.

The variance detector and decision support agents provide earlier warning by surfacing problems like reserve shortfalls and return-condition gaps while there's still time to act. Rather than stopping at data extraction, the pipeline delivers insights through decision support, which transforms the findings into something executives can actually use to evaluate extensions, buyouts, or portfolio moves.

Getting started

These deployment steps walk you through standing up the quickstart and running a document through it.

Where this goes

NeIO LeasingOps relieves human experts of the document grind—reading, reconciling, monitoring deadlines—so they can use that time for strategy and relationships instead. This quickstart lets you try this in your own Red Hat OpenShift AI environment to see whether an agentic system fits your organization’s workflow.

Keep learning

• GitHub

• Contact: For deployment help, email [email protected] or [email protected] .

• License: The Helm chart and deployment configuration are Apache 2.0. The application images are proprietary.

Block | Dynamic pattern deluxe promo

Component | Band_header

Product trial

Red Hat OpenShift AI (Self-Managed) | Product Trial

An open source machine learning (ML) platform for the hybrid cloud.

Component | spacer

Component | Cta_multi_basic

Subpattern | simple_cta

Component | CTA

Try it

Deluxe mbox

Component | Card_header

About the authors

Subpattern | speaker

Card layout

Component | Image_embed

Component | Person

Michael Dawson

Node.js lead for Red Hat and IBM

Subpattern | social_links

Michael Dawson is an active contributor to the Node.js project and chair of the Node.js Technical Steering Committee (TSC). He contributes to a broad range of community efforts including platform support, build infrastructure, N-API, Release, as well as tools to help the community achieve quality with speed (e.g., ci jobs, benchmarking and code coverage reporting). As the Node.js lead for Red Hat and IBM, he works with Red Hat and IBM's internal teams to plan and facilitate their contributions to Node.js and v8 within the Node and Google communities. Dawson's past experience includes building IBM's Java runtime, building and operating client facing e-commerce applications, building PKI and symmetric based crypto solutions as well as a number of varied consulting engagements. In his spare time, he uses Node.js to automate his home and life for fun.

More from this author

Bertrand Rault

Senior Principal Engineering Partner Manager

As a Senior Principal Engineering Partner Manager at Red Hat since 2016, he specializes in Kubernetes and AI technologies across both hardware and software. His expertise spans Red Hat OpenShift, Red Hat OpenShift AI, Red Hat AI Inference, and Red Hat Enterprise Linux AI. He works closely with strategic technology partners to co-develop integrated solutions, accelerate innovation, and bring new technologies to market. His focus is on turning complex technology ecosystems into practical, integrated solutions that deliver value to customers.

Bala Sista

Head of Technology and Architecture

• Bala Sista is the Head of Technology and Architecture at Codvo.ai, where he shapes the company’s technology strategy and oversees its product portfolio, led by NeIO, Codvo’s agentic AI platform. He leads solution architecture across enterprise engagements in energy, manufacturing, healthcare, telecom, and financial services.

With 29 years of experience, Bala began his career working with industrial control systems on factory floors, followed by 15 years developing data historian products at Invensys and Schneider Electric. His work resulted in eight US patent filings, seven of which were granted. He has also held executive engineering roles in AI products and platforms, including assistive AI for blind and low-vision communities at Aira and personal AI at Personal.ai.

At Codvo, Bala sets technological direction, leads architecture for client engagements, and has built the company’s AI-native engineering practice, including agent-driven development, review gates, certification frameworks, and delivery tools. He continues to architect enterprise AI systems and review technology deliverables.

Indranil Paul

Director – Product & Solutioning, Codvo.ai

Indranil Paul is Director – Product & Solutioning at Codvo.ai, where he leads product strategy, solutioning, and enterprise AI and data transformation initiatives. With over 18 years of experience across AI/ML, Agentic AI, Vision AI, data platforms, product strategy, and engineering, he helps organizations convert emerging technologies into scalable solutions and measurable business outcomes.

His work spans NeIO—Codvo’s enterprise Agentic AI platform—as well as partner technologies and clients’ existing AI, data, and enterprise ecosystems. He leads engagements from identifying high-value use cases and developing solution proposals to creating business cases, evaluating ROI, shaping transformation roadmaps, and supporting commercialization and go-to-market execution.

Indranil has helped shape industry-specific solutions across energy and LNG, industrial operations, financial services, aviation leasing, healthcare, and manufacturing. An MBA Gold Medalist from IIM Bangalore and a certified SAFe Product Owner, he is passionate about applying AI responsibly to transform business processes, augment human expertise, and create sustained enterprise value.

More like this

Dynamic pattern

Blog post

Agentic AI for network operations: Smarter decisions on proven automation

Balancing open science and data privacy with enterprise Red Hat OpenShift AI

Original podcast

How Red Hat cleared IT debt for scalable AI

Standardizing the AI stack with PyTorch

Subpattern | card_flex

Subpattern | text_basic

Keep exploring

• What is agentic AI? Article

• Predictive AI vs. generative AI Article

• Top considerations for building a production-ready AI/ML environment E-book

• Generative AI, the Ansible way Video

• Innovate and transform with a modern application platform E-book

Keep Exploring mbox

Subpattern | simple_text

Browse by channel

Explore all channels

Pattern | raw_html

Automation

The latest on IT automation for tech, teams, and environments

Artificial intelligence

Updates on the platforms that free customers to run AI workloads anywhere

Open hybrid cloud

Explore how we build a more flexible future with hybrid cloud

Security

The latest on how we reduce risks across environments and technologies

Edge computing

Updates on the platforms that simplify operations at the edge

Infrastructure

The latest on the world’s leading enterprise Linux platform

Applications

Inside our solutions to the toughest application challenges

Virtualization

The future of enterprise virtualization for your workloads on-premise or across clouds