I work on browser testing tools and LLM reliability (open to work). 2026 CS graduate, Delhi.
10 PRs merged: 8 into Google's Puppeteer and Chrome DevTools MCP, 1 into Hugging Face, 1 into Confident AI's DeepEval, plus bugs I found in Microsoft Playwright, Twenty and Documenso that their teams fixed.
- Puppeteer · #15587 WebSocket handshake timeout · #15590
setGeolocation(null)clears the override · #15595respond()skips undefined headers on CDP · #15697 BiDirespond()/abort()no longer reject for gone requests - Chrome DevTools MCP · #3144 · #3148 · #3198 · #3199 console, network and Lighthouse fixes
- sentence-transformers · #4204 PListMLE custom rank discount received a mask instead of ranks
- DeepEval · #3470 JSON dataset loader crashed on null
tools_called/expected_tools(reported as #3469) - Playwright · reported #43459, a codegen escaping bug, fixed in #43472
- Documenso · reported #3445, ~26 redundant
waitForTimeoutcalls (~27.5 s of fixed sleep) found by parsing the E2E suite with the TypeScript Compiler API, fixed in #3536 - XERJ Corpus Hub · my #3199 fix is indexed as a contributor corpus, a reference library AI coding agents search before writing code
- Twenty · reported #28342, Links field dropping slashes in query/hash, fixed in #28421
| Junban | Finds the API response order that breaks a flaky Playwright test. Caught 6/6 rare races by run 5; 3,000 plain reruns caught only 2/6. |
| Zeitaku | Replaces hard-coded sleeps in Playwright tests with network-aware waits over CDP. Built after the Documenso #3445 analysis. |
| HalluGuard | Flags RAG answer sentences their sources don't support, using an NLI cross-encoder. |
| Kensa | Metamorphic testing for RAG apps. |
| DocMind | Question answering over PDFs, scanned ones included, with hybrid search. |
| Attention-Guard | Learns your personal "drifting" pattern from activity and 20-minute check-ins, and nudges you when focus slips. Runs fully on-device. |
TypeScript · Python · Playwright · Chrome DevTools Protocol · PyTorch
Writing: I keep building tools that catch software lying about whether it works
nancy6547.github.io · LinkedIn · X · Medium · singhnancy2004@gmail.com

