A RAG-powered QA assistant grounded in ISTQB Foundation Level, IEEE 829, and the Software Testing Life Cycle (STLC) — plus a companion artifact evaluator that scores any QA deliverable against weighted rubrics.

🏆 Winner — Codecademy Agentic AI Applications Bootcamp Contest #1 (2026)

🔗 Live demo: ambitqa.com 🤖 Chatbot: ambitqa.com/chatbot.html 🔬 Evaluator: ambitqa.com/evaluator.html

Chatbot currently at v1.4.4, Evaluator at v1.3.11 — see Changelog.

Need QA work done faster, with documentation that passes any audit?

I use AmbitQA as my personal force-multiplier to deliver senior-grade QA artifacts in a fraction of the typical time — grounded in ISTQB Foundation Level, IEEE 829, and modern STLC practices. Available for short engagements and ongoing retainers.

• ✅ Real QA practitioner, not a generalist consultant
• ✅ Tool-augmented delivery — what would take a junior QA 2 weeks, I deliver in 3 days with higher quality
• ✅ Methodology-grounded output — every artifact references ISTQB / IEEE 829 sources, audit-ready by default
• ✅ Transparent scoping — fixed fees on defined deliverables, no surprises
• ✅ Code stays yours — all artifacts and frameworks are delivered as your property

• 💼 LinkedIn: linkedin.com/in/hkapur
• 📧 Email: hitesh@ambitqa.com
• 📅 Book a 30-minute intro call: calendly.com/hiteshkapur/30min

Currently accepting up to 2 new engagements per month. Best fit: startups and mid-size teams needing senior QA capacity without a full-time hire.

AmbitQA is a structured, retrieval-augmented chat agent that produces production-grade QA deliverables in seconds — not generic AI chat output. Every response is grounded in a curated methodology knowledge base and cites the exact sources used.

A separate evaluator scores any QA artifact (whether produced by AmbitQA or copy-pasted from anywhere) against weighted rubrics, returning a PASS / WARN / FAIL verdict with dimension scores, top issues, and an auto-generated improvement prompt.

• Test plans (IEEE 829 format)
• Risk assessments and risk registers
• Test cases — tabular and BDD/Gherkin
• Bug reports — full IEEE 829 defect structure
• Requirements Traceability Matrices (RTM)
• Test execution summary reports
• Exit criteria checklists
• Lessons-learned documents
• Automation framework code — Playwright (TypeScript), Cypress (JavaScript), Selenium WebDriver (Python + pytest), and BDD/Cucumber + Gherkin
• Plus risk-based strategies, accessibility checklists, exploratory test charters, API contracts, security test suites, performance test scenarios, and more

• Curated knowledge base with methodology chunks (ISTQB, IEEE 829, STLC, automation patterns, OWASP, k6/JMeter, REST/GraphQL)
• Semantic keyword scoring + mode-aware retrieval boost
• Top-K relevant chunks injected into every prompt
• Source citations on every response — auditable, not hallucinated

A standalone artifact scoring engine (evaluator.html) that:

• Evaluates any QA artifact against 13 weighted rubrics (test plan, risk assessment/register, test cases, BDD, bug report, automation code, RTM, execution report, exit criteria, lessons learned, API test cases, security test cases, performance test plan/script)
• Returns PASS / WARN / FAIL verdict with dimension-level bar charts
• Lists the top issues found
• Generates an improvement prompt you can paste back into the chatbot
• Pre-scans for red flags before scoring
• Integrates with the chatbot via the 🔬 Evaluate button on every response, or the standalone 🔬 Evaluate Artifact button in the nav bar (for scoring artifacts written anywhere, not just ones the chatbot produced)
• Multi-provider on its own terms — same 5 providers as the chatbot, with per-provider keys remembered separately; launching from the chatbot pre-selects whichever provider/model was active there
• Drop a .txt/.md file, click to browse for one, or paste directly — and a ✕ Clear button resets the artifact, file selection, and any previous results in one click

Works with multiple LLM providers — switch on the fly:

• Anthropic (Claude Sonnet 4.6 and other models)
• OpenAI (GPT-5.6 family)
• Google (Gemini 2.5 / 3.5)
• xAI (Grok 4.3 and newer)
• OpenRouter (any model on their platform)
• Custom endpoint — bring your own OpenAI-compatible API

Per-provider API keys stored separately (session-only in the chatbot; remembered across sessions in the Evaluator). Custom model and endpoint fields for any provider.

Each mode has phase-specific quick actions and RAG retrieval boost:

• All Phases
• Requirements Review
• Test Planning
• Test Design
• Test Execution
• Defect Management
• Test Closure
• Test Automation (Playwright / Cypress / Selenium / BDD)
• API Testing (REST, GraphQL, Postman, JWT/OAuth)
• Security Testing (OWASP Top 10, XSS/injection/IDOR, Burp/ZAP)
• Performance Testing (k6, JMeter, load/stress/spike/soak)

• Set a target URL → agent fetches the page through a dedicated Cloudflare Worker proxy (with public fallback proxies as backup), avoiding the rate-limit flakiness of relying on third-party free proxies alone
• Extracts forms, buttons, navigation, and structure from the HTML
• Generates site-specific deliverables referencing actual UI elements
• Heavily bot-protected sites may still decline the fetch — falls back gracefully to URL + description mode, or attach the page's HTML manually for guaranteed results

Three attachment paths to give the agent maximum context:

• Screenshots (PNG/JPG/WEBP/GIF) — vision capability sees the UI directly
• PDFs — native PDF understanding
• HTML files — DOM structure extracted and injected for real, working selectors in automation code

Attach via 📎 button, drag-and-drop, or paste from clipboard.

A dedicated STLC phase that generates production-ready code:

• Playwright POM + tests + playwright.config.ts + README
• Cypress POM + spec files
• Selenium WebDriver framework (Python + pytest)
• BDD Gherkin features + Cucumber.js step definitions
• GitHub Actions CI workflow
• Honest about selector sources — comments tell you whether selectors came from real HTML or were inferred from screenshots

• TXT — clean plain text (markdown stripped, tables formatted as records)
• MD — raw markdown (great for GitHub, Notion, Obsidian)
• HTML — standalone styled document (one click away from PDF via browser print)
• Jira CSV — direct-importable for test cases and bug reports
• Auto-form fill — when the agent asks for structured input, a modal opens with proper input fields

Cancel in-flight responses via the Stop button (AbortController). Useful when an output is going in the wrong direction.

• Three-column resizable layout with drag handles
• Both side panels can collapse for full-screen chat
• Light & dark mode (defaults to system preference)
• Indigo/violet brand palette
• WCAG AAA-compliant text contrast
• Mode-aware autosuggestions in the chat box (↑↓ navigation, Enter to pick)
• Scroll-to-top / scroll-to-bottom navigation buttons that track the chat column's actual layout (survive panel resizing and toggling)
• 🔑 Change Key button — re-enter or fix a key for the active provider without switching away and back
• Version badge driven by a single constant for clean release management
• Keyboard-friendly throughout

Pure browser-based — no backend, no database, no server-side app code (one small Cloudflare Worker handles target-page fetching only).

• Static HTML + vanilla JS + CSS (chatbot.html ~3,150 lines, evaluator.html ~1,240 lines)
• Hosted on GitHub Pages
• Calls Anthropic / OpenAI / Google / xAI / OpenRouter / Custom APIs directly, each in its native request format
• A dedicated Cloudflare Worker fetches target-page HTML server-side (bypassing browser CORS restrictions) with public proxies as fallback
• API keys entered in-browser; chatbot keys are session-only, Evaluator keys are remembered per-provider for convenience — never sent to any server other than your selected provider's API
• Multi-modal request format (images, PDFs as binary; HTML as extracted text context)
• 48K token output ceiling with truncation detection + continue button

+--------------------------------------------------------------+
|  Browser (the entire app)                                    |
|                                                              |
|  +------------------+    +-----------------+                 |
|  | Knowledge base   |    | STLC mode       |                 |
|  | (curated chunks) |    | (11 modes)      |                 |
|  +--------+---------+    +--------+--------+                 |
|           |                       |                          |
|           v                       v                          |
|  +-----------------------------------------+                 |
|  | Semantic retrieval (top-K by mode boost)|                 |
|  +--------------------+--------------------+                 |
|                       |                                      |
|                       v                                      |
|  +-----------------------------------------+                 |
|  | Prompt builder + multi-modal content    |                 |
|  | (text + images + PDFs + HTML context)   |                 |
|  +--------------------+--------------------+                 |
|                       |                                      |
|                       v                                      |
|         api.anthropic.com / api.openai.com / etc.            |
|              (selected provider + model)                     |
|                       |                                      |
|                       v                                      |
|  +-----------------------------------------+                 |
|  | Markdown renderer + export pipeline     |                 |
|  | (TXT . MD . HTML . Jira CSV)            |                 |
|  +--------------------+--------------------+                 |
|                       |                                      |
|                       v                                      |
|  +-----------------------------------------+                 |
|  | Evaluator (rubric-based scoring)        |                 |
|  | PASS / WARN / FAIL + improvement prompt |                 |
|  +-----------------------------------------+                 |
+--------------------------------------------------------------+

1. Visit the chatbot
2. Pick a provider (Anthropic, OpenAI, Google, xAI, OpenRouter, or Custom)
3. Paste your API key for that provider (used only in this browser session)
4. Pick an STLC mode from the left panel
5. (Optional) Set a target URL in the right panel for site-specific deliverables
6. (Optional) Attach screenshots or HTML files for visual/structural context
7. Use a quick action or type your prompt — start typing to see autosuggestions
8. Export the result as TXT, MD, HTML, or Jira CSV
9. Click 🔬 Evaluate on any response to score it

1. Visit the evaluator directly, or launch it from the chatbot via the 🔬 Evaluate button on any response, or the standalone 🔬 Evaluate Artifact button in the nav bar (for scoring an artifact you wrote yourself, or from anywhere else)
2. If launched from the chatbot, the provider/model are pre-selected to match — just add a key for that provider if you haven't already
3. Paste any QA artifact (test plan, risk assessment, test cases, bug report, etc.)
4. Select the artifact type (or leave on Auto-detect)
5. Get instant verdict, scores, issues, and improvement prompt

• Chatbot: your API key is used only in the browser session — never persisted, never sent anywhere except your selected provider's API
• Evaluator: keys are remembered per-provider in browser localStorage for convenience across visits — a deliberate tradeoff for a tool you'll reuse repeatedly; clear your browser storage to remove them
• Target URL fetching routes through a small Cloudflare Worker (operated by the author) that fetches the page server-side to work around browser CORS restrictions — it sees the target URL you enter, not your API key or conversation content, and doesn't log or store anything
• No analytics, no tracking, no cookies
• Code is fully open-source — audit it yourself
• All requests use explicit browser-access opt-ins required by each provider

Recommendation: Set per-key spending caps in your provider's console for any keys used in browser-based tools.

• AI: Anthropic, OpenAI, Google, xAI, OpenRouter (any model, switchable)
• Frontend: Vanilla HTML / CSS / JavaScript (no framework, no build step)
• Hosting: GitHub Pages
• No backend, no database, no dependencies

Chatbot and Evaluator now version independently (they ship on their own schedules).

v1.4.4

• Quick actions now tag the artifact type they produce (e.g. "Generate a full IEEE 829 test plan" → Test Plan), passed to the Evaluator via sessionStorage and embedded as metadata in TXT/MD/HTML exports — replaces guessing the type from output text after the fact with a deterministic signal from generation time

v1.4.3

• Added "Clear all" button to the attachment preview row (shown once there are 2+ pending attachments)

v1.4.2

• Evaluator launches pre-selected to the chatbot's active provider/model

v1.4.1

• Updated default model IDs for Google, xAI, and OpenAI (previous defaults had been deprecated by their providers)

v1.4.0

• Fixed multi-provider support — previously every request silently went to Anthropic regardless of the selected provider; now properly routes to each provider's native format, including image/PDF handling per provider's actual capabilities

v1.3.x

• Fixed scroll-to-top/scroll-to-bottom navigation buttons through several root causes (DOM timing, viewport-vs-column positioning, panel-toggle staleness) — now tracks the chat layout reliably via ResizeObserver
• Added dedicated Cloudflare Worker for target-page fetching, replacing sole reliance on free public CORS proxies
• Added standalone 🔬 Evaluate Artifact button in the nav bar
• Added 🔑 Change Key button for the active provider
• Added Risk Assessment as a first-class chatbot deliverable type

v1.3.11

• Reads artifact-type metadata (from chatbot, via sessionStorage or embedded file markers) before falling back to keyword-based auto-detect — a manual dropdown selection always overrides both

v1.3.10

• Added automatic HTML-export detection and stripping — pasting the contents of an exported .html file (instead of the plain markdown/text response) now correctly extracts the visible content before auto-detect and scoring, instead of choking on the raw CSS/markup

v1.3.9

• Follow-up fix for auto-detect misclassification (v1.3.8's frequency-weighting helped but wasn't sufficient on its own) — now also heavily weights keywords found in the document's title/first heading, since an explicit self-declared title is a far more reliable signal than word frequency in body prose

v1.3.8

• Fixed auto-detect misclassifying documents based on incidental single-word matches (e.g. a Test Plan scored as Bug Report) — now weights keyword frequency instead of mere presence
• Fixed evaluations failing with a JSON parse error on some models (response was hitting the token limit mid-output) by raising the output ceiling and improving the error message for this case

v1.3.7

• Added ✕ Clear button to reset the artifact text, file selection, type dropdown, and any previous results in one click

v1.3.6

• Added click-to-browse file upload (drop zone previously only accepted drag-and-drop)

v1.3.5

• Added multi-provider support (previously Anthropic-only), mirroring the chatbot's mechanism, with per-provider key memory and pre-selection when launched from the chatbot

v1.3.4

• Added Risk Assessment / Risk Register rubric (previously these were misclassified into an unrelated rubric by keyword auto-detect)

v1.3.3

• Added API Test Cases, Security Test Cases, and Performance Test Plan/Script rubrics (9 → 12 rubrics)

• Stop generation button (AbortController)
• Version badge driven by a single constant
• Custom model field with persistence
• Custom provider flow with URL validation gating
• Multiple UX fixes for evaluator and chatbot

• Multi-provider support (Anthropic, OpenAI, Google, xAI, OpenRouter, Custom)
• Three new STLC modes: API Testing, Security Testing, Performance Testing
• Companion evaluator with 9 weighted rubrics
• Evaluate button (🔬) on every response toolbar
• Nav restructured

• Initial release as Codecademy bootcamp capstone
• 8 STLC modes, RAG knowledge base, multi-modal input, 5 export formats

• Heavily bot-protected sites — the dedicated Worker proxy handles most sites reliably, but aggressive anti-bot protection (e.g. major e-commerce/search platforms) may still decline the fetch; falls back to URL + description mode (workaround: attach HTML file or screenshot)
• Token limits — very large outputs (~48K tokens) may be truncated; a Continue button resumes the response
• No conversation persistence — refreshing the page clears history (intentional for privacy)
• Browser API key — anyone with access to your browser can see keys entered; chatbot keys clear when the tab closes, Evaluator keys persist in localStorage until manually cleared (see Privacy & security)

Copyright © 2026 Hitesh Kapur. All rights reserved.

This code is published for portfolio demonstration and educational reference.

You may:

• ✅ View, read, and study the source code
• ✅ Reference patterns or approaches in your own original work
• ✅ Use the live demo with your own API key
• ✅ Share links to this repository and the live site

You may NOT:

• ❌ Copy, redistribute, or republish this code or any substantial portion
• ❌ Deploy this code (in whole or in part) to any public or commercial service
• ❌ Use this code in any commercial product or paid offering
• ❌ Remove attribution or claim authorship

See LICENSE for full terms.

For commercial licensing inquiries, please contact via LinkedIn (linked above).

The methodology knowledge base references (ISTQB Foundation Level concepts, IEEE 829 templates, STLC definitions, OWASP Top 10, k6/JMeter patterns) are based on publicly available industry standards and are not claimed as original works.

• Codecademy for the Agentic AI Applications Bootcamp — and for awarding AmbitQA the Contest #1 winner spot
• Anthropic for Claude and the multi-modal Messages API
• OpenAI, Google, xAI for their respective LLM APIs
• The ISTQB, IEEE, and OWASP communities for the methodology foundations this is built on

Built by Hitesh Kapur — find me on LinkedIn · #CodecademyAgenticAIBootcamp