Use Cases
Vedana is built for domains where answer correctness matters more than “fluency of text”. Below are the typical scenarios where it outperforms classic RAG.
E-commerce and catalogs
What it can do
- Answer questions about specific products with current prices and stock;
- filter by attributes (“all frost-free fridges under 100k”);
- compare products by attributes;
- check availability at specific branches;
- map products to regulatory requirements (category → requirements → required certificates).
Why classic RAG isn’t a fit
Prices, stock, and characteristics are structured data that change. Vector search will find old mentions, not actual values. Vedana, via Cypher, goes directly to the graph and returns current values.
Example
“Which euro-book sleeper sofas under £600 do you have in stock at the Nevsky store?”
→ Vedana builds Cypher with filters product.mechanism="euro-book", product.price < 600, (p)-[:PRODUCT_available_at_BRANCH]->(:branch {name: "Nevsky"}), and returns the full list with prices and SKUs.
Legal and compliance
What it can do
- Which regulations / requirements apply to a product / contract / category;
- which documents regulate which area;
- contract dates and deadlines;
- approval statuses and blockers;
- multi-hop reasoning across documents and regulators.
Example
“Which documents do we need to show the customer when selling built-in appliances in the EU?”
→ Vedana traverses product → category → regulation → required_documents and returns the requirements list with citations from the documents.
Internal knowledge bases
What it can do
- search policies, regulations, instructions;
- answer org structure questions (“who is responsible for X”, “who do I talk to about Y”);
- explain processes with citations from the source;
- support employee onboarding.
Example
“What’s the approval process for an equipment purchase budget over 500k?”
→ Vedana finds chunks of the regulation + walks the graph “process → stages → owners”, then gives a step-by-step answer with a quote from the document.
B2B and dealer support
What it can do
- stable answers to typical dealer questions;
- export specs, price lists, partner terms;
- product compatibility checks, configurator;
- routing requests to the right manager.
Example
“Can product X be bundled with Y, and what would the final price be for our dealer tier?”
→ Vedana checks compatibility through (x)-[:COMPATIBLE_WITH]-(y) and computes the price including the dealer tier through a structured query.
HR and org structure
What it can do
- who on the team works on which project;
- who’s responsible for a process / system / customer;
- who reports to whom and who to escalate to;
- schedules, vacations, on-call rotations.
Example
“Who handles the integration with customer Foo, and who is their sales manager?”
→ (:client {name: "Foo"})-[:HANDLED_BY]->(:person)-[:REPORTS_TO]->(:person) — Vedana returns two people with contacts.
Finance and reporting
What it can do (with care)
- answers about reference info (policies, limits, procedures);
- structured queries about deal statuses, contracts;
- helping navigate reports.
What you shouldn’t do
- compute revenue directly through the LLM. Numbers come from BI/DWH through manual dashboards or via a custom tool.
Less suitable scenarios
Vedana won’t replace:
- Creative writing. Vedana is rigidly tied to data; the LLM has little freedom.
- Free-form summarization. If the task is “read it and tell me in your words”, classic RAG is simpler.
- Pure conversational AI without a domain. With no structured data and complex questions, Vedana is overkill.
- Code / API doc search. Better to use specialised tools (e.g. Cursor / GitHub Copilot / Cody).
- Real-time analytics on huge volumes. Memgraph is analytical, but it’s still a graph database, not Clickhouse.
A template for evaluating a new case
Before taking on a case, answer three questions:
- Can the answer be expressed as a structured query or a graph traversal? If yes, Vedana fits best.
- Does the answer live in text documents, and does it need meaning-based search? If yes, Vedana via vector search + chunking.
- Can a combination of both be used? If yes, that’s the best fit for Vedana.
If all three are “no”, look at other tools.