Custom AI Development
For MSP Partners and Their Clients

When Standard Prototyping Is Not Enough

The Petronella Fleet prototyping ladder handles the majority of MSP engagements. A PoC Lite at $35,000 proves feasibility. An MSP Fleet Prototype starting at $50,000 delivers a working system with a bill-of-materials and handoff documentation. Most regulated-SMB deals fit cleanly into one of those tiers.

But some client requirements push beyond standard prototyping into genuine custom AI development work:

• Multi-model orchestration. The client's workflow requires multiple specialized models coordinated through an agent framework, not a single inference endpoint. A legal review pipeline might route documents through a classification model, an extraction model, a summarization model, and a compliance-check model in sequence, with branching logic at each stage.
• Custom fine-tuning. Off-the-shelf models produce adequate but not excellent results on the client's domain-specific data. Fine-tuning on proprietary datasets creates a model that understands the client's terminology, document formats, and decision patterns in ways that prompt engineering alone cannot achieve.
• Domain-specific RAG at scale. The client has hundreds of thousands of documents, regulations, or records that need to be searchable through natural-language queries with citation-grade accuracy. The vector database design, chunking strategy, retrieval pipeline, and re-ranking logic all require custom engineering beyond what a standard prototype delivers.
• Compliance-grade inference pipelines. The client operates in a regulated environment where every inference must be logged, auditable, and reproducible. ITAR-compliant document processing, HIPAA-grade medical record analysis, or CMMC-scoped defense manufacturing workflows all require purpose-built pipelines with compliance controls baked into the architecture, not bolted on afterward.

For these engagements, Petronella Technology Group delivers custom AI development as an extension of the Fleet services-only model. We build on our private fleet, document the architecture, train your MSP team on operations, and hand off a working system. Your client procures their own hardware using the bill-of-materials we deliver. Your MSP manages the deployed system for the client under your own contract.

What Custom AI Development Means at Petronella

This is not application development in the traditional sense. Petronella's custom AI work focuses on infrastructure and model-layer engineering:

The distinction matters for MSPs: Petronella does not build web applications, mobile apps, or SaaS products. We build the AI infrastructure layer that applications connect to. If your client also needs a user interface, dashboard, or integration layer, that work is scoped separately and may involve your own development team or a third party.

Examples of Custom Work

These are representative engagement types, described generically to protect client confidentiality. They illustrate the scope and complexity of custom AI development work:

Private LLM for Legal Document Review

A regulated firm needs to review thousands of contracts, extract specific clauses, flag non-standard terms, and produce summaries for attorneys. The model must run entirely on-premises with no cloud connectivity. The pipeline ingests documents in multiple formats, normalizes them, runs extraction through a fine-tuned model, and outputs structured data with page-level citations. The MSP manages the deployed hardware and the inference endpoint; Petronella builds the pipeline and trains the MSP team on operations.

Medical Image Analysis Pipeline

A healthcare organization needs to process diagnostic imaging data through a classification model that flags anomalies for radiologist review. The pipeline must comply with HIPAA technical safeguards, log every inference for audit purposes, and integrate with the facility's existing PACS system. The architecture includes data ingestion controls, model serving with role-based access, audit logging, and an output interface that feeds results back into the clinical workflow.

ITAR-Compliant Document Processing

A defense manufacturing client needs to process technical documents that contain controlled unclassified information under ITAR restrictions. No data can transit cloud services, all processing must occur on authorized hardware, and the system must produce audit trails that satisfy DFARS 7012 preservation requirements. The pipeline includes document classification, entity extraction, redaction automation, and secure storage with chain-of-custody logging.

The Handoff Model

Custom AI development follows the same services-only philosophy as the rest of the Petronella Fleet program. Petronella does not sell hardware, does not rent multi-tenant infrastructure, and does not create ongoing dependencies that bypass the MSP.

The handoff sequence works like this:

1. Petronella builds. All development and prototyping happens on Petronella's private fleet. Your client's data is processed under NDA with appropriate security controls for the data classification level.
2. Petronella documents. Every engagement produces an architecture document, runbook, bill-of-materials for production hardware, deployment guide, and operations manual. Documentation is written for the MSP's technical team, not for Petronella's internal use.
3. Petronella trains. Your MSP team receives hands-on training on the deployed system: how to monitor, how to troubleshoot, how to update models, and how to respond to performance degradation. Training is scoped in the statement of work and delivered before the handoff is considered complete.
4. MSP manages. After handoff, your MSP operates the system for the client under your own managed-services agreement. The client writes checks to you. Petronella's direct involvement ends unless the MSP opts for ongoing Managed Service support.

This model matters because it preserves the MSP's economic position. You are not renting Petronella's infrastructure indefinitely. You are paying for engineering work that results in a system your MSP owns the operational relationship for.

Why Services-Only Matters for Custom AI

Hardware procurement is the MSP's responsibility in every Fleet engagement, and custom AI development is no exception. Petronella Technology Group does not sell GPUs, servers, or networking equipment. We do not take hardware margins, carry inventory, or assume warranty exposure.

For custom AI work specifically, this means:

• The bill-of-materials is vendor-neutral. Your MSP can procure from NVIDIA, Supermicro, Dell, CDW, or any other source.
• There is no lock-in to Petronella's infrastructure after the prototype phase. The system runs on hardware the client owns.
• Petronella's pricing reflects engineering time, not hardware markup. This keeps the services margin clean for both Petronella and the MSP.
• If the MSP wants help managing the procurement process, the optional $2,500 procurement coordination fee covers vendor selection, specs, purchase orders, and delivery tracking.

Fleet Tiers That Map to Custom Work

Custom AI development engagements enter the Fleet ladder at the higher tiers where scope and compliance requirements justify the engineering investment:

• Compliance-Aware MSP Prototype at $75,000 is the entry point for most custom AI work. The 4-to-6-week timeline covers a working prototype with CMMC, HIPAA, or NIST 800-171 compliance mapping built into the architecture from day one.
• Production-Ready MSP Prototype from $125,000 is for engagements that require a signed migration guarantee, 90-day post-handoff support, and an embedded CMMC-RP on the MSP's Slack channel during transition. This tier is inquiry-only and invoiced via wire or ACH after custom scoping.

Both tiers gate through a free 15-minute Discovery Call. No Stripe Payment Link is sent before the call because custom work scope varies too widely for self-serve pricing to be appropriate.

How Custom Engagements Are Qualified Before Any Money Moves

A five-figure custom AI engagement should be qualified as carefully by the MSP as by the vendor. The qualification gate on both sides is the same: a free 15-minute Discovery Call, and no payment link is issued before it. Here is what the call needs to establish, from your side of the table:

• The use case, stated as a workflow. "Review incoming contracts and flag non-standard indemnification clauses" is a workflow that can be prototyped and measured. "We want to use AI more" is not yet an engagement. If your client cannot name the workflow, the right first step is usually a smaller conversation, not a bigger tier.
• The data environment. What formats, how many documents or records, where they live today, and what classification governs them. Custom RAG and fine-tuning work depends entirely on the corpus; a vague answer here produces a vague prototype.
• The compliance overlay. CMMC Level 2, HIPAA, NIST SP 800-171, or none. This single answer moves the engagement between the standard tiers and the Compliance-Aware tier, because the compliance artifacts are deliverables that have to be engineered, not appended.
• Who owns the client relationship. Fleet and custom development are services-only and co-managed: your MSP invoices the client and holds the relationship, and Petronella Technology Group delivers engineering under your SOW. Deal registration under the free Petronella Partner agreement protects your opportunity for 90 days.
• The production path. What happens after the prototype works. Hardware budget, deployment window, and who operates the system. An engagement with a production plan is a better fit for the higher tiers than an exploratory one.

Bring those five answers and the tier recommendation on the call is grounded in scope rather than in budget. If the honest answer is that the engagement is not ready, we say so on the call.

What the Compliance Overlay Adds to an Inference Pipeline

For partners whose clients operate under CMMC Level 2, HIPAA, or NIST SP 800-171, the compliance layer is not paperwork around the system; it is engineering inside it. Understanding what that layer contains helps an MSP set the client's expectations before the Discovery Call:

• Access control mapped to the client's roles. The inference pipeline inherits the client's identity model. Which users can query which corpora, and which outputs they may act on, are enforced in the pipeline rather than assumed in the application.
• Audit logging of every inference. A full audit trail from query to response. This is what makes an AI system defensible rather than merely private.
• Data boundary and retention controls. Network isolation, encryption at rest, and retention rules stated in the statement of work. The engagement retention window defines when prototype data is destroyed; production retention is designed into the deployed system.
• Documentation artifacts. The Compliance-Aware tier and above produce compliance mapping, SSP artifacts, and an audit-evidence package. These sit alongside the rest of the client's compliance documentation, so the AI system is governed by the same program as everything else in the environment.
• Change management. Model updates, prompt changes, and pipeline modifications go through controlled change processes after deployment. A system that drifts silently is a finding waiting to happen, so the managed-service layer treats model changes the way change control treats any other system change.

None of this converts Petronella Technology Group into an assessor: Petronella Technology Group, Inc. is a Cyber AB Registered Provider Organization (#1449) with a CMMC Registered Practitioner team, which means we help clients and MSPs prepare for and support assessments, and documents signed by a credentialed practitioner always carry the Petronella Technology Group, Inc. name.

How the MSP Prices and Positions the Deployed System

Process guidance only: one way to structure the commercial layer around a custom AI engagement. Your pricing to your client is yours to set.

One-time layer: the engagement itself

The prototyping engagement is a project with defined deliverables. Partners generally present it to the client as the de-risking step for a capital decision: the client pays for a validated architecture, a bill-of-materials, and compliance artifacts, not for hardware sight unseen. Because the fee is paid upfront and non-refundable, scope clarity on the Discovery Call is what protects both sides.

Recurring layer: operations

The durable economics are in operating the system: monitoring, patching, capacity management, model updates, and compliance-aware change management. The Managed Service stage prices this for Petronella Technology Group under your SOW, and your MSP marks its own layer on top for the client. Partners who scope this layer before the client demo tend to convert one-time projects into recurring relationships.

Expansion: the second and third use case

A deployed private AI system is a platform. Additional use cases, additional departments, fine-tuning refreshes, and agent builds all reuse the foundation of the first engagement. The architecture document and workload profile from the handoff are the documents that make those follow-on engagements straightforward to scope.

For MSPs Wanting an Ongoing Custom Development Relationship

If your MSP has multiple clients with custom AI needs, or a single large client with a multi-phase AI roadmap, the Petronella Strategic Partnership may be a better fit than individual Fleet engagements. The Strategic Partnership provides ongoing access to Petronella's engineering capability under a retainer structure rather than per-project pricing.

The Strategic Partnership is application-only and designed for MSP owners with $8M or more in annual revenue and a demonstrated pipeline of regulated-SMB clients who need private AI infrastructure. It includes full access to the Petronella capability stack, including custom AI development, CMMC Registered Practitioner bench, and DFE forensics, delivered by the team under a single retainer.

Related Capabilities Within the MSP Partners Program

• Petronella Fleet for the full 4-tier prototyping ladder and services-only engagement model
• Petronella Strategic Partnership for MSPs wanting an ongoing advisory and engineering relationship
• CMMC assessment practice for defense-sector clients whose custom AI work must satisfy CMMC Level 2
• HIPAA compliance practice for healthcare clients requiring compliance-grade inference pipelines
• Apply to the MSP Partners program to begin the scoping process
• MSP Partners program hub for the full 4-tier partner ladder overview

Frequently Asked Questions