01
Extract entities
The NLP module parses structure: parties, amounts, dates, requisites, annex links, and regulatory references.
partyamountdaterequisite
On-prem · AI document reviewWe turn your expertise
We turn your expertise
into executable rules for an
AI assistant.
Review documents against your internal policies and checklists inside your security perimeter.
contract_AB-7724.pdf · page 14 / 287
analysis
3.2 Subject of contract
The supplier undertakes to transfer equipment specified in Specification No. 2 dated 14.03.2026, to the buyer's ownership, total contract value
contract value is stated in figures and words, including VAT 20%.
Delivery term — no later than 30 business days from the signing date.
3.3 Warranties
The warranty period is 6 months from acceptance certificate signing.
Entities
Amountmatch
VAT20%
Term30 b.d.
Warranty6 mo.
Checks
Amount in words matches
Specification No.2 details
Warranty < 12 mo. (by checklist)
Clarify liability for delay
-80%
review workload
90%
checks automated
3,000
docs per day
2-3 mo
pilot to go-live
Problem
Document volume grows faster than the team.
Contracts, reports, and internal policies keep growing while hiring and onboarding specialists takes months. In banking and exchanges, mistakes cost fines and reputation.
1-4 hours
of manual review per document
A lawyer or methodologist checks each document against the checklist, switches between dozens of files, and copies findings into spreadsheets.
20-30%
misses and regulatory risks
Fatigue, different interpretations of rules, and hidden risks in 50+ page attachments lead to inconsistent review quality.
x2-x3
growth that hiring cannot absorb
Document flow grows faster than the team. Hiring and onboarding take months, while regulatory deadlines do not wait.
How it works
From document to report in 4 steps inside your security perimeter.
No file leaves your environment. Self-hosted LLM, RAG, and vector storage run on your infrastructure or in your VPC.
02
Run the checklist
Your checklist becomes executable: every rule is checked by the LLM and validated with document quotes.
rule_4.2ok
rule_4.3ok
rule_5.1fail
03
Find discrepancies
Cross-checks across package documents, contracts, policies, text clauses, and regulatory references.
critical3
attention7
ok37
04
Generate the report
A structured report with evidence quotes, DOCX/PDF export, and redline format for legal teams.
DOCXPDFJSON
Use cases
Six scenarios, one engine.
SemantIQ adapts to your department methodology: from currency control to non-standard contracts. Your team keeps its current process; SemantIQ follows your rules.
case - large financial institution
Currency control
Review of foreign trade contracts: currency, delivery basis, repatriation, compliance with regulator instructions.
JTBD
So that a currency control specialist can close contracts within the required 1-3 business days without losses under regulatory reporting.
Effect
-91% time
Before
90 min
After
8 min
Ways to deploy
Three ways to use SemantIQ in your workflows.
01 / Web
Web UI
Upload a document package, navigate findings, and export the report. Built for analysts and methodologists.
02 / Plug-in
Office plug-in
Review directly in Word: redline comments in the margin, quote highlighting, and a 'Run checklist' button.
Word · Document.docxSemantIQ
risk
Note:Warranty period < policy threshold (12 mo.)
03 / API
Backend pipeline
REST/Kafka workflow for 3,000 documents per day with DMS, BPM, and core-system integrations.
# submit package to SemantIQ
POST /v1/check
{
"checklist": "fx_control",
"docs": ["..."],
"async": true
}
→ 202 accepted · job_18a7c
Architecture
Deployed inside your security perimeter.
No external APIs. Everything from the LLM to vector storage runs on your hardware or in your VPC.
Customer security perimeter
Documents
PDF · DOCX · XLSX
NLP
Entity extraction
parsing, NER, links
LLM
Self-hosted model
7B / 70B · GPU on-prem
RAG
Context search
policies, regulations
Vector DB
Vector database
document embeddings
Rules
Checklists
versions, role model
Glossary
Glossary and regulations
internal terminology
Output
Structured report
- Findings + quotes
- DOCX redline
- JSON for BPM/DMS
- Audit log
On-prem
in your data center or VPC
Self-hosted LLM
data stays in your environment
Audit, roles, logs
SSO, RBAC, full audit trail
Checklist versioning
git-style diff, rollback
Rollout
From pilot to production in 8 weeks.
From kickoff to a production-ready workflow without moving data outside your security perimeter.
Phase 01weeks 1-2
Onboarding
Checklist setup, access, environment deployment, and pilot document set upload.
Kick-off, scope
On-prem deployment
Golden-set collection
Phase 02weeks 3-4
Calibration
Checklist-to-rules mapping, methodologist iterations, and entity extraction tuning.
Rule mapping
LLM calibration
Interim metrics
Phase 03weeks 5-6
Parallel run
SemantIQ works in parallel with the team; we compare results and tune accuracy.
Shadow-mode
Discrepancy analysis
Rule refinement
Launchweeks 7-8
Production go-live
Production workflow, user training, handover to operations, and SLA.
Production workflow
Team training
SLA and support
Impact calculator
Estimate how much manual work can be removed from the team.
Move the sliders. The calculation shows workload before rollout and after automated review with short expert validation.
Assumptions: after rollout, a specialist spends about 5 minutes validating a document, while SemantIQ performs the main checks against your checklist.
5020 000
10240
Before SemantIQ
1,500 h
per month on manual review
After SemantIQ
83 h
per month on expert validation
Freed / month
1,417 h
Freed / year
17,000 h
Automation
94%
Commercial terms
Pricing is tailored to your deployment.
Final terms depend on document volume, number of workflows, on-prem deployment requirements, and integrations. Leave a request and we will prepare a quote for your setup.
Base
up to 500 doc/day
Start for one department
On request
One scenario, Web UI, basic SLA. A fast entry point for the first team.
- 1 checklist / department
- Web UI + DOCX/PDF export
- SSO, basic audit
- Support 8x5
Minimal setup without unnecessary integrations
Discuss pilotRecommended
Business
up to 1,500 doc/day
Optimal for several teams
On request
2-3 scenarios, Office plug-in, extended SLA. For regular flow and several processes.
- Up to 3 checklists
- Web UI + Office plug-in
- RBAC, checklist versioning
- Support 24x5
- DMS / BPM integration
Best balance of rollout speed and coverage
Discuss pilotEnterprise
3,000+ doc/day
Maximum control
On request
Unlimited scenarios, dedicated team, custom integrations and models for your regulations.
- Unlimited checklists
- Backend pipeline + API
- Custom models and integrations
- Support 24x7, dedicated TAM
- Joint R&D roadmap
Private AIAPI24x7
Contact usWhy ReML
Nine years of applied AI engineering for large enterprises.
ReML is an independent AI integrator. We do not sell a magic button — we build production systems that pass security audit, methodologist review, and real user workflows.
9+
years in the market
200+
AI cases in production
20+
projects in 2025
80%
repeat contracts
In 8 weeks, the pilot moved to a production pipeline for currency control. SemantIQ removes up to 80% of routine checks from the team and consistently catches what previously required manual review weeks later.
MB
Business line owner
Large financial institution
Case
Currency control
Team
Core team
Product, architecture, and delivery leaders who take pilots from methodology to production workflows.
Yaroslav Shmulev
CEO, founder ReML
- 9+ years of ML delivery and Data Science consulting
- Ex ML Lead at SAP
- Expertise in CV, predictive analytics, and modern AI
- Scientific publications
Stas Okrug
CPO
- In AI since 2016: successful ML, CV, GenAI products
- Expertise in SOTA technologies
- MIPT graduate (with honors) and Yandex School of Data Analysis
- Lecturer at MIPT, Sber University, and DLSchool
Dmitry Pirozhenko
CTO
- 8+ years in backend development and information security
- Architecture of high-load AI systems
- Lecturer at Innopolis University in information security
FAQ
Frequent questions from security and methodology teams.
Did not find an answer? Write to us — we will send a whitepaper and deployment scheme under NDA.
Where does the model run and where does data go?
The model and all components are deployed on-prem or in your VPC. Documents and embeddings never leave the customer perimeter. External APIs are not used.
What SLA do you provide for availability and processing time?
Standard SLA is 99.5% availability and typical document processing under 90 seconds. Enterprise contracts can include 99.9% and dedicated infrastructure.
How are checklists updated?
Checklists are versioned with branches and rule diffs. A new version can be tested on a golden set before production rollout. All changes are logged.
What do you do with our data for training?
Nothing without separate written consent. Client data stays on the client side. Fine-tuning is done only on synthetic or anonymized data by agreement.
How long does rollout take?
A shadow-mode pilot takes 8 weeks. Production go-live usually takes another 4-8 weeks depending on integrations such as DMS, BPM, and core systems.
Which systems do you integrate with?
REST/Kafka, typical DMS and BPM systems, Active Directory / SSO, and internal buses. The Word plug-in is available for users who work with documents manually.
We will show it in 30 minutesTest it on your documents
Test it on your documents
in 1-2 weeks.
Send 5-10 documents and your current checklist. In 1-2 weeks you will see results on your own corpus — without integrations or migrations.
info@re-ml.com