RAG pilot on a corporate basis
The pilot tests the corporate knowledge base on real documents, users, and questions prior to full-scale deployment. There are two fixed-duration editions: "Lite" lasting four weeks and "Full" lasting six weeks.
Login Format
The pilot answers a question that cannot be verified on a demo stand: whether the assistant provides useful responses specifically to your documents and your employees' questions. We take one process, a limited set of sources and real users—and evaluate answer quality, not general impressions.
If the sources contain personal data, contractual terms, or trade secrets, access and masking rules must be defined before indexing, not after the first complaints.
What's included
Scenario
We select one process and its owner, define the user group and criteria by which we ultimately decide: was the pilot successful or not.
Sources
We examine which documents and regulations will be included in the database, who is responsible for their relevance, and how access rights to originals are transferred to the assistant's responses.
Pilot environment
We are building an index and search, configuring responses with links to sources, enabling query logging, and collecting user feedback.
Quality assessment
We are preparing a set of test questions, experts annotate the answers, we analyze the errors and document what needs to be improved before the production launch.
If the pilot result needs to be formally accepted—with thresholds, a protocol, and signatures—the measurements can be carried out on a test bench. PolygonIt checks the load and response quality in a single run and issues a PASS/FAIL verdict based on agreed criteria.
If the RAG must not only respond, but also create tasks, launch processes, or access an API, the pilot contacts protected by AI automation: action restrictions, logs, approvals and human review.
Editions and volume limits
The term is chosen before the contract and sets the duration: four weeks or six. After signing, the term is not an open range.
| Parameter | Lite — 4 weeks | Full — 6 weeks |
|---|---|---|
| Case | up to 300 documents | Up to 2,000 documents, no more than 10,000 pages, and 2 GB of source files |
| Sources | one source with file export | up to two sources with file export |
| Users | up to 10 | up to 20, up to three access groups, up to five simultaneous sessions in the test run |
| Quality check | shortened test set, one setup cycle | 100 test questions, two cycles of search and answer refinement |
| Meetings | up to four working meetings | up to six interviews and working meetings |
Common to both editions: one process, one Russian-language corpus, one environment and one web interface, one aligned LLM, and one embedding model. Document formats: PDF with text layer, DOCX, TXT.
What is needed from the client
- the process owner and experts who will prepare reference answers;
- corpus cleared of information prohibited for use in the pilot;
- access matrix and approved data processing mode;
- agreed computing resources and model access — prior to the start of work;
- Time for experts to annotate and provide feedback: up to 20 hours total in the "Full" edition.
What is not included
- scanning and complex table recognition, manual cleaning of the entire corpus;
- Custom connectors to ERP and ECM, model training;
- AI agent actions in external systems;
- SSO and full-fledged IDM, complex dynamic permissions at the individual record level;
- industrial fault tolerance and SLA, full-scale load testing, certification.
Project materials are transferred within the scope of rights established by the contract. If a pilot is assembled on Ragify, this does not imply transfer of rights to the platform itself or its perpetual free use—licensing terms are specified separately.
When the work is completed
Technical acceptance criteria and the utility hypothesis are separate — they are different things.
- Mandatory technical handover: A fully configured and approved test environment has been deployed; the corpus has been loaded and indexed; search, answers with source references, logging, and approved access restrictions are operational; results are reproducible with fixed versions of the corpus, model, and settings.
- Quality check: A test was conducted on an agreed set of questions, results and error analysis were provided, and the configuration cycles specified by the editorial team were completed.
- Utility hypothesis: The target threshold for valid responses is set before launch. Failure to meet it allows the conclusion "do not scale," but does not waive mandatory technical criteria.
Access checks on a limited set do not guarantee the security of an industrial solution. The conclusion "do not scale" is a valid pilot outcome: it saves implementation budget.
Output documents
| Artifact | Why is it needed? |
|---|---|
| Working pilot | Expanded RAG pipeline on the selected process and limited set of sources, available to pilot group users |
| Quality Assessment Report | Results on the test set of questions: proportion of useful answers, common errors, completeness of source references |
| Safety requirements | Roles and access rights, handling of sensitive data, logs, and restrictions that need to be migrated to the production environment |
| Scaling plan | What is needed for production launch: source volume, integrations, load, support, and the sequence of department connections |
Along with the documents, a functioning test stand is provided for an agreed period (by default—up to 30 calendar days after handover, provided the customer's resources are maintained), as well as a deployment diagram and instructions, a registry of sources with an access matrix, a test dataset with reference annotations, a registry of information security constraints, and all configurations and scripts created within the project. The basic test protocol is included in the pilot; extended measurements on the test stand Polygon —a separate option, not the only way to obtain formal acceptance.
Package budget
| Editorial office | Package budget |
|---|---|
| Lite — 4 weeks | 700,000 – 1.2 million ₽ excluding VAT |
| Full — 6 weeks | 1.8 – 3 million ₽ excluding VAT |
The exact fixed price is determined after the initial meeting and scope verification; included work, deliverables, and exclusions are confirmed prior to commencement; any scope expansion is possible only upon agreement.
Platform licenses, computing resources (GPU or cloud), and the use of external models are billed separately and displayed prior to the agreement.
Team composition
- AI and RAG Engineer — index, search, answer quality, handling source links.
- Data Engineer — connecting sources, preparing documents, updating the database.
- Information Security Architect — access rights, handling of sensitive data, logs.
- Business Analyst and Project Manager — process selection, test questions, working with a pilot group, result evaluation.
Frequently asked questions
Can I use Ragify?
Yes. Ragify is suitable as a product basis for a RAG pilot.
Do you need all company documents?
No. It's best to start with a single process and a limited set of sources.
What if there is not enough data?
The pilot will show you what sources are needed and how to improve your knowledge base.
How we fix the volume and deadline
The terms below are identical for all packages and are incorporated into the contract without changes.
- Perimeter — prior to the agreement. We hold an initial meeting of up to 60 minutes and agree on the perimeter passport: systems, processes, sources, participants, and outcomes.
- Countdown of the term. The timeline starts after the handover of the approved set of source materials, provision of access credentials, and confirmation of the interview schedule. Delays on the client's side will postpone the timeline by a corresponding amount of time.
- Data. We operate within a coordinated secure channel. Production data is not transmitted to external AI services without the customer's written permission.
- Result transmission. We provide materials in PDF and editable formats (DOCX, XLSX, PPTX, or agreed equivalents) and conduct one results defense session of up to 90 minutes.
- Acceptance. Five business days are allocated for reviewing the results. One consolidated round of clarifications within the agreed scope is included; however, correction of errors and non-compliances with the contract is not limited to this round.
- Volume change. New systems, sources, and tasks are included only upon agreed changes to scope and cost.
- Work format. The default format is remote, unless an in-person component is explicitly planned for the package.
The statuses and licenses on the basis of which the works are performed are collected on the page "Licenses and statuses".
RAG pilot - second stage implementation of AI in business processes: after examination and before industrial implementation.
RAG pilot - the first stage of the project: development of a chatbot with AI for business.
Email us
Write to us at info@restart.re or click the button below. For the first letter, it is sufficient to briefly describe the outline, the role of the organization, current systems, limitations and the desired outcome.
Email usRequest format
We agree on the volume, timing and output artifact - and launch it in a short time.
