Enterprise Cognitive AI
Domain intelligence on infrastructure you control
LoopQ.si runs domain-tuned language models, agents and your own ML models on your servers or your private cloud. Every answer is benchmarked on your data, grounded in your documents and governed end to end.
Documents, scans and tables in; OCR for scanned pages
Knowledge bases and graphs over your own corpus
Open-weight models fine-tuned on your data, beside your own ML
Visual multi-step agents for real workflows
Every model and agent gets its own audited API
The challenge
Why public AI APIs stall in regulated enterprises
Data leaves your perimeter
Prompts carrying policy wordings, claims files and customer records are processed on someone else's infrastructure, outside your control and your audit trail.
Fluent, unverified answers
General models answer confidently about exclusions, limits and underwriting rules they were never grounded in, and nothing checks them before a customer sees them.
Integration lands on IT
A raw open model leaves your team to provision GPUs, tune serving, wire in documents and keep it all patched: skills that are scarce and costly to hire.
The platform
One platform from raw document to governed answer
LoopQ.si is Model-as-a-Service built on self-hosted, open-weight models. The model comes to your data, not the other way round.
Data Hub
Ingest policy wordings, manuals, contracts and claims files. Scanned pages go through OCR.
Knowledge
Knowledge bases and knowledge graphs built over your own corpus.
Domain SLMs
Open-weight small language models fine-tuned on your data, served beside your own ML models.
Agent Studio
Compose models, knowledge and tools into multi-step agents from visual blocks.
Governed API
Each model and agent is published with its own API, keys and audit trail.
Runs on
Your on-premises GPU servers, your private cloud, or managed GPU capacity we operate for you.
Across every stage
Evaluation, guardrails, access control, encryption and a complete audit trail.
Building models
Domain SLMs that speak your business's language
Small language models, typically 7 to 14 billion parameters, cost less per answer than frontier models and fit on a single on-premises GPU. Fine-tuned on your documents, they learn the terms your people actually use.
Learns your language
Fine-tuned on your policy wordings, manuals and case history, so it uses your jargon and acronyms. No hand-built dictionaries or grammars.
Recipes, not data scientists
Base model, datasets, training settings and the quality gate are packaged as a recipe your team reruns from the portal as your data grows.
Visual agent workflows
Agent Studio composes models, knowledge bases and tools into multi-step agents, so teams prototype and test without heavy coding.
Bring your own: PyTorch, scikit-learn and Hugging Face models run beside your SLMs, under the same governance.
Internal Model-as-a-Service
Your AI team publishes models. Every department uses them.
Central AI team
Builds, evaluates and approves models and agents, with our engineers alongside.
LoopQ.si
- Catalogue of approved models and agents
- One governed API per model and agent
- Keys, quotas and usage metered per project
Call it like any AI service. REST or SDK, the way teams already call third-party APIs, but it runs on your hardware.
Cost lands where it's used. Usage is metered per project, so each department sees what it consumes.
Access is explicit. Role- and attribute-based access control, with every call logged for audit.
Infrastructure and security
Dedicated infrastructure is a control, not a lock-in
Data residency
Models, documents and indexes run on hardware you choose: on-premises or in your own cloud tenancy, in the country you choose. No external model provider sits in the request path.
Encryption and access
Per-tenant encryption keys managed through a dedicated key-management service. Role- and attribute-based access control, scoped to the project, on every model, dataset and agent.
Dedicated capacity
No public rate limits and no noisy neighbours. GPU capacity is sized to your workload and measured on your hardware during the pilot. Our contractual commitment is availability.
Air-gap ready
Installs from container images onto your own GPU servers, including NVIDIA DGX-class systems. Once model weights are loaded, inference needs no outbound connection.
Built to support data-protection and technology-risk obligations on where data is processed and who can reach it, from Malaysia's PDPA and BNM RMiT to regional equivalents in India, the Gulf, the UK and the EU.
Answer quality
Every answer is tested, grounded and accountable
Benchmarked on your cases
Models are scored on your own scenarios against a weighted reference rubric, by an evaluator model you also host.
Gated releases
Version 2 is benchmarked side by side with the live model on the same cases, and is promoted only if it doesn't regress.
Grounded and cited
Answers cite the passages they drew on, and each claim is checked against the cited source text.
Monitored and accountable
Guardrails and drift monitoring in production, signed receipts per prediction, and Risk, Compliance and Executive sign-off.
We measure the hallucination rate on your data and show it to you. We don't promise zero; we prove the number.
Where it works
Built for document-heavy, regulated work
Insurance and takaful
- Check claims against the policy wording
- Underwriting rule validation
- Policy and benefits Q&A for agents and staff
Banking and financial services
- Fraud and AML signals
- Credit and capital-risk models
- Regulatory document review
Enterprise operations
- Invoice and purchase-order matching
- Supplier and spend-risk scoring
- Knowledge search across manuals and contracts
Pricing
A transparent commercial structure
One flat platform subscription. Compute is your choice and is sized from real measurements in the pilot.
Platform
per month
- Platform licence and every update
- 10 hours of expert engineering support each month
- Data Hub, knowledge bases, Domain SLMs, Agent Studio
- Benchmarks, quality gates, guardrails and audit
Compute
sized in the pilot
- Cloud GPU, pay as you go
- Or a dedicated on-premises node, bought as fixed capex
- Or bring GPU servers you already own
Start
low-friction pilot
- One high-value workflow, scoped together
- Isolated sandbox with your sample documents
- Accuracy measured on your own cases
Prices exclude applicable taxes (such as SST, GST or VAT). Prices in currencies other than MYR are list prices for that market and are confirmed on your quote.
| Area | Build in-house | LoopQ.si |
|---|---|---|
| People | Hire and retain scarce AI and ML infrastructure engineers | Flat subscription with 10 expert engineering hours included |
| Time to value | Typically 6 to 9 months to set up hardware, model serving and data pipelines | Pre-configured platform, with a working pilot within 30 days |
| Upkeep | Your team owns model updates, drift and security patches | Platform updates, drift monitoring and patching included |
| Governance | Evaluation, audit trail and approvals built from scratch | Benchmarks, quality gates, audit and approvals built in |
Expert support
Ten expert hours a month, spent on your workflows
We don't hand you a login and a documentation link. Our engineers work directly with your team.
Domain tuning
Work with your experts to build training data, fine-tuning recipes and prompt templates.
Quality loops
Build benchmarks, scoring rubrics and review queues where your assessors label answers.
Workflow build-out
Configure agents for claims parsing, underwriting checks and document review, and connect them to your systems.
Next steps
From vision to production: the 30-day pilot
Select the scope
Pick one high-value manual document workflow, such as checking claims against a policy wording.
Set up the sandbox
We stand up an isolated environment and load your sample documents and test cases.
Validate the model
The first 10 expert hours benchmark answers on your cases and review the scores with your team.
Get in the loop
Drop your datasets to in@loopq.si and get your team in the loop.
Send a sample of the documents behind the workflow you want to automate, such as policy wordings, claims forms or manuals. We'll come back with what a pilot on them would look like.
FAQ
Questions we hear most
Is LoopQ.si a wrapper around ChatGPT or another public AI service?
No. LoopQ.si serves open-weight models on infrastructure you control, alongside your own ML models. Your prompts and documents are not sent to a third-party model provider.
Where does our data live?
On the hardware you choose: your own servers, your cloud tenancy in your chosen region, or dedicated capacity we operate for you. Encryption keys are per tenant, and every access is logged.
Do we need a data-science team?
No. Fine-tuning runs from packaged recipes, agents are built from visual blocks, and the 10 monthly expert hours cover the specialist work with your domain experts.
What hardware do we need?
That depends on the models and volume, which we measure during the pilot. Options range from pay-as-you-go cloud GPUs to a dedicated on-premises node such as an NVIDIA DGX-class system, or GPU servers you already own.
How do you handle hallucination?
We measure it rather than promise it away. Models are benchmarked on your cases before release, answers cite their sources and each claim is checked against them, and a quality gate blocks any new version that does worse than the one in production.
What does the service-level agreement cover?
Availability is our contractual commitment. Response times depend on the hardware and workload, so we measure them on your hardware during the pilot instead of quoting a generic number.
Which currencies do you invoice in?
We quote in MYR, INR, USD, GBP, EUR, JPY, AED, OMR, BHD, SAR, QAR, KWD and SGD. Prices exclude local taxes, which are added on your invoice where they apply.
Contact
Book a scoping session
Tell us about the workflow you'd like to automate. We'll set up a short call and, if it fits, your secure sandbox.
Email hello@loopq.si