The matching engine behind Sogiry
Comparing prices is the easy part. Knowing two listings are the same product is the hard part.
Across Lithuania, Latvia, and Estonia, no two stores describe the same product the same way. Sogiry reads millions of listings in three languages and works out which are genuinely identical, which are close-but-different variants, and which should never be compared — the product graph that makes safe, automated repricing possible.
- 4M+ listings analysed
- Three languages, one product graph
- Every match is explainable
Two stores, same box
Sony WH1000XM5 belaidės ausinės, juodos
LT · no SKU
Sony WH-1000XM5B Wireless Headphones, Black
EN · SKU WH1000XM5B
Resolved product
Sony WH-1000XM5
Black · Wireless headphones
94%
match confidence
One product, a hundred descriptions.
If every store used the same title and the same product code, price comparison would be a spreadsheet. Real Baltic ecommerce looks nothing like that — and closing that gap is the whole job.
Every store speaks differently
The same headphones are “Sony WH1000XM5 juodos” in one shop and “Sony WH-1000XM5B Wireless” in another — different language, word order, units, and abbreviations describing the exact same box.
Variants change everything
A colour, bundle, package size, warranty, or region code can turn a perfect match into the wrong one. Matching the product but missing the variant means bad prices, so we treat sub-products as first-class.
The market never holds still
Products vanish, URLs redirect, sellers relist, and competitor catalogues change daily. A static list of competitor URLs is stale the moment it is built — our graph re-checks itself continuously.
Five stages, each one you can inspect.
We deliberately break matching into separate, observable steps. That keeps the system improvable — and keeps pricing from ever becoming a black box.
Collect the signals
Titles, prices, categories, stock status, seller details, and how each offer behaved over time.
Clean and structure
Messy titles become clear facts — brand, model, size, colour, package — the same shape for every store.
Compare and classify
Embeddings, classifiers, and rules separate exact matches from look-alikes and unsafe comparisons.
Build the product graph
Listings group into products and sub-products, each with a confidence score and its evidence.
Decide within guardrails
Graph signals become repricing suggestions, bounded by margin floors, approvals, and an audit trail.
The AI proposes. Your guardrails decide.
A matching engine that quietly changes prices is a liability. Every match Sogiry makes is scored, explained, and reversible — and nothing reaches a price tag without passing your rules.
Confidence, not guesses
Each match carries a score. High-confidence matches can auto-apply, uncertain ones go to review, risky ones are excluded — you set the thresholds.
Every match shows its work
See exactly why two listings were matched: shared attributes, price behaviour, seller data, and where variants differ. No mystery.
Pricing stays under control
Minimum margins, floor prices, approval workflows, pause switches, and a complete audit log sit between intelligence and action.
4M+
Listings analysed
A growing dataset of Baltic catalogues and seller offers.
3
Languages, one graph
Lithuanian, Latvian, and Estonian product data resolved into a single model.
SKU-level
Decision granularity
Signals resolve to products, sub-products, and repricing-safe actions.
24/7
Continuous refresh
The graph re-checks itself as offers, stock, and prices move.
A product intelligence layer for the whole region.
Repricing is the first application. The deeper opportunity is a living, domain-specific understanding of how products, sellers, variants, and prices move across the Baltics — and how AI shopping assistants answer questions about them.
Book a technical discussion- 01Expand the product graph across Lithuania, Latvia, and Estonia.
- 02Deepen multilingual product and variant extraction.
- 03Strengthen retrieval and similarity search for merchant catalogues.
- 04Connect graph signals to AI shopping visibility and automated pricing.