Test automation at scale,
now with AI.

Systematic, multi-judge evaluation of LLM outputs with hallucination detection & regression tracking

Replaces subjective "does this look right?" prompt testing with structured, repeatable LLM evaluation. YAML test suites run against Groq-hosted models, scored by parallel LLM judges with hallucination detection and regression tracking — visualised in a live Streamlit dashboard you can run directly from the browser.

• YAML-driven test suites — hallucination, safety, format, and RAG evaluation
• Multi-judge consensus (Groq + Cerebras) with disagreement detection
• Hallucination detection via claim extraction + batched NLI verification (~60% fewer tokens)
• Relevance scoring using sentence-transformers cosine similarity
• Regression tracking with automatic score drop detection across runs
• Run live evaluations from the browser — no CLI required
• Jira auto-ticket creation on consecutive failures

Requirements → approved test plans → AI execution → Jira triage — with a human in the loop

Takes unstructured requirements and converts them into executed, reported, and triaged test cases end to end. A RequirementAnalyst agent generates JSON test plans; a Streamlit portal lets humans approve before any test runs. Failures trigger AI bug triage and auto-open Jira tickets with screenshots.

• RequirementAnalyst agent: plain-text requirements → structured JSON test plans via LangChain LCEL
• Human-in-the-Loop Streamlit portal — approval gate before execution
• LLM provider failover: SambaNova → Groq → Ollama (one config line)
• Parallel execution via pytest-xdist; Allure reports with per-failure AI analysis
• ActionRegistry replaces page-object models — Playwright driven by AI decisions

Ask questions across your QA docs — test cases, bug reports, SRS — and get grounded answers with source attribution

A production-grade Retrieval-Augmented Generation system built from scratch to understand every layer of RAG. Documents are chunked, embedded locally (no GPU needed), and stored in Qdrant Cloud. At query time the user's question is embedded, matched against stored vectors, and the top chunks are passed to Groq for a grounded answer. No black boxes — every component is explicit and replaceable.

• LangChain RecursiveCharacterTextSplitter — splits on paragraphs → sentences → words for clean chunk boundaries
• sentence-transformers (all-MiniLM-L6-v2) — free, CPU-only, 384-dim embeddings
• Qdrant Cloud vector DB — cosine similarity search with source filename + score in every result
• Groq (llama-3.1-8b-instant) — ~1s grounded answers; returns "Not found in QA docs" when context is absent
• Streamlit chat UI with session history, collapsed source expander, and in-app re-indexing

Zero-script bug detection & test case generation via LLM-guided browser crawl

An AI testing agent that crawls any web application, detects bugs, and auto-generates test cases without a single manually written selector. The LLM decides what to explore; Playwright executes it. Supports Groq (cloud) and Ollama (fully local).

• Auto-login detection — Cloudflare, multi-step login, SSO, cookie banners
• Signal-gated bug detection — LLM only fires on real errors (console, XHR, DOM)
• Self-healing actions with up to 5 fallback strategies on selector failure
• API testing layer captures all XHR/fetch calls during crawl and validates each endpoint
• Generates contextual test cases per page saved to Excel; Allure reports auto-open