Levana Test Assistant

A domain-aware QA copilot that turns multi-chain trading behaviour into structured, selector-aware, automation-ready test coverage.

The Challenge

Levana's trading experience changes across chains, wallets, environments, and live market states. Generic test generation could not reliably capture wallet sequencing, chain-specific behaviour, trading risk, error states, or the precision needed for downstream automation.

System Workflow

01Grounded generated cases in real Levana trading flows rather than generic prompts
02Adjusted scenarios for Osmosis, Injective, Neutron, supported wallets, and testnet or mainnet behaviour
03Covered connect, trade, order, wallet rejection, balance, margin, latency, and state-synchronisation paths
04Added concrete UI and selector references to reduce ambiguity
05Structured output for Playwright and Java-based automation workflows
06Used LangSmith to trace prompt context, chain and wallet decisions, generation steps, and evaluation results

Human Controls

QA review keeps generated scenarios aligned with current product behaviour

Domain rules and known user journeys constrain what the assistant produces

Automation-ready output is reviewed before it becomes executable coverage

LangSmith traces and evaluations make generation paths inspectable before QA approval

Impact & Evidence

01Standardised test-case quality across complex multi-chain trading flows

02Reduced the manual effort needed to design and maintain validation coverage

03Improved the path from exploratory testing to repeatable automation

04Supported broader release-readiness testing without removing QA ownership

05Added trace-level visibility for investigating weak outputs and comparing future prompt or model changes

Artifacts

Related Project Case Study

View the Levana product and QA case study

Explore the wider multi-chain product context, testing challenge, and delivery outcomes.