Catalog scale
thousands of SKUs
The problemEnriching a large catalog and covering support peaks makes cost the deciding factor.
HelmcodeA flat rate per key: the whole catalog and every campaign at a fixed, predictable cost.
industries · E-commerce & retail
The whole catalog, every campaign, one rate.
Catalogs of thousands of SKUs and seasonal support peaks make cost the deciding factor. Semantic search, product content and support on a flat rate, measurable in conversion.
the economics
The economics, by design.
Each cost lever mapped to a platform capability. Sovereignty comes by default.
Conversion
semantic search
The problemOn-site search quality moves conversion more than almost anything else.
Helmcodeqwen3-embedding plus reranking: the natural entry case, and measurable in conversion.
Sovereignty
EU · GDPR
The problemCustomer chats and data must stay GDPR-compliant.
HelmcodeEU-only inference with zero logs for the customer-facing chatbot.
how search quality is measured
Why there is no best embedding model.
On-site search is the lever that moves conversion, and it runs on an embedding model. The public benchmark for those models reached a conclusion that should shape how you buy: there is no winner to pick, only a fit to find.
01
Eight tasks, not one
The Massive Text Embedding Benchmark spans 8 embedding tasks over 58 datasets and 112 languages. Retrieval, which is what a catalogue search does, is one task among several, and a model tuned for classification or clustering is being measured on something else.
02
Nobody dominates
Having benchmarked 33 models, the authors state plainly that no particular text embedding method dominates across all tasks. The field has no universal answer, which means a vendor telling you they have the best embedding model is describing one column of a wide table.
03
So keep the swap cheap
That conclusion is an argument about architecture rather than about any model: what pays off is being able to re-embed your catalogue with a different model when the evidence changes, in your own language and your own domain, without renegotiating anything.
MTEB · Massive Text Embedding Benchmark "MTEB: Massive Text Embedding Benchmark", arXiv:2210.07316, with a public leaderboard that has kept growing since. The figures above are the benchmark as published; the leaderboard ranking changes month to month, which is rather the point of the third finding. read the report →
use cases
Your most common use cases.
The cases with the most traction in the sector, each with its own page in detail.
Semantic search & recommendation Catalog semantic search, the entry case, measurable directly in conversion. Content generation Product descriptions and category copy at catalog scale. Customer support Customer support that absorbs seasonal peaks under GDPR. Visual inspection Product photos checked against your own catalogue rules, and returns triaged from the picture the customer sent. Classification & triage Catalogue categorisation and support routing across thousands of SKUs, at a cost that does not scale with them. Conversation QA Post-peak review of every support conversation, which is when you find out what the campaign actually broke.
Recommended open models.
A starting point per task type. The full guide maps 80 cases to the open model for each one.
- qwen3-embedding + rerank in Helmcode
- Catalog semantic search and ranking: the conversion lever.
- Qwen3.6-27B
- Product descriptions at volume with good writing on one GPU.
- DeepSeek V4 Flash in Helmcode
- Customer support and classification at seasonal-peak volume.
We are distilling and quantizing these open models into small, tightly specialised versions, trained for one task rather than for all of them. A model like that runs on less hardware, answers faster and fits where the big one does not, your own datacenter included. If you have a process with volume and stable criteria, that is the conversation we want to have with you.
// faq
Questions, answered.
What the sector's technical, compliance and business teams ask.
What is the natural first use case?
Catalog semantic search: it improves the shopping experience and its impact is measurable directly in conversion, which makes it easy to justify.
Can it handle the whole catalog and campaigns?
Yes. A flat rate per key removes the per-token ceiling, so enriching thousands of SKUs and running every campaign costs the same fixed amount.
Is the customer chatbot GDPR-compliant?
Prompts are never stored (zero logs) and inference runs only on EU infrastructure, so the customer-facing chatbot stays GDPR-compliant by architecture.
How does it integrate?
The API is OpenAI-compatible: change the base URL and key and your storefront, PIM or search stack keeps working unchanged.
// get started
START BURNING TOKENS
Skip the AI infra work. Deploy your first private inference endpoint today.
Flat rate. EU data. OpenAI API compatible.