Vertical AI · Milan
AI that works on your premises, by your rules.
V-AI builds V_RAG, an edge-first hybrid AI system for regulated companies: language models run at your site and the cloud steps in only when needed. Fewer tokens, less energy, data under control.
The problem
Four obstacles hold back generative AI in regulated companies
Personal data
Case data should not reach third-party models.
Per-token cost
It grows with the documents: a long file takes more calls and tens of thousands of tokens.
Cloud dependency
A risk that financial-sector rules require companies to manage.
Errors on rules
General-purpose models get specific rules wrong and invent articles.
The solution
The local model does the work, the cloud steps in only when needed
V_RAG brings AI to where the documents are. For each request V_Router decides whether the local vertical model is enough or a cloud model is needed, and then sends it only requests without personal data, with reduced context.
- V_Boxlow-power appliance with key and signed releases
- V_LLMcompact language models, vertical by sector
- V_RAG enginereads PDFs, email and certified email, retrieves sources, verified citations, hash-chained log
- V_Metertokens, energy and CO2e measured per case
- Cloud modelonly requests without personal data, with reduced context
- Coordinationseparate per customer: signed updates
V_LLM
Compact open models, specialised on the sector and on the customer's documents.
V_Router
Local or cloud for each request, with cloud fallback. It also works with the local model alone.
V_Meter
Tokens, energy and CO2e per case in a dashboard, also useful for the customer's reporting.
Designed around the rules
An architecture designed for the requirements our customers must meet
Data minimisation and privacy by design: documents stay at the customer, only requests without personal data go to the cloud.
AI suggests, the operator decides. Hash-chained log and citations verified against the sources.
Human decision and transparency, built into the product.
ICT third-party risk: the system keeps working with the local model alone.
Supply-chain security: signed releases and controlled updates.
Measured targets
Fewer tokens and less energy per case
of requests served locally, at equal accuracy
tokens sent to cloud models per case
energy and emissions measured for each case by V_Meter
Targets of the development project, to be verified in customer pilots.
Who it is for
Companies that handle many cases under written rules
Insurance, health funds, BPO
Claims, reimbursements, complaints, access to records.
Legal and compliance
Checking documents against regulations, contracts and procedures.
Administrative services
Completeness and verification of case files.
Typical customer: 300 to 5,000 cases a month.
Getting started
From pilot to contract
Process analysis
1-2 weeksTogether we choose the cases, the rules and the indicators.
Pilot on your data
8-12 weeksAgreed KPIs: accuracy, minutes saved, tokens, energy.
Annual contract
recurring feeOne-off activation and verticalisation, fee by case volume tier.
Contact
Let's talk about your process
We are selecting companies for the first pilots. Write to us: we answer personally.
- Contact person
- Paolo Osvaldo Agnelli, founder
- poa@v-ai.services
- Phone
- +39 335 212640
- Office
- Piazzale Biancamano 2, 20121 Milano