South Asian guy brewing zeros and ones.
Computational Social Science · Applied AI · Public Policy
I'm a master's student at IIT Jodhpur studying Computational Social Science. My research is on digital inequality and AI governance. Outside of that I write software — mostly small security tools for AI agents — and send fixes upstream when I find bugs in libraries I use.
Merged upstream fixes:
| Project | Fix |
|---|---|
| fairlearn/fairlearn (2.3k★) | #1801 — stop deprecated pandas Series growth in reductions |
| mlco2/codecarbon (1.9k★) | 3 merged PRs, incl. #1439 — preserve multi-digit GCP region numbers |
| python-babel/babel (1.5k★) | #1401 — keep timedelta sub-second precision through format_timedelta |
| petl-developers/petl (1.3k★) | #766 — close XLSX workbooks when iteration exits early |
| pypa/pyproject-metadata | #345 — preserve empty inline readme and license text |
| xarray-contrib/cf-xarray | #674, #675 — coordinate and geometry decoding fixes |
| sara-czasak/py-simple-wrap | #402 — streaming CSV row iterator for easy_csv |
- agent-config-audit — scans AI agent configs for risky permissions, unpinned MCP servers, exfiltrating hooks and prompt-injection patterns
- agent-threat-model — describe an LLM-agent system in YAML, get a STRIDE + OWASP Agentic threat model with a ranked register
- agent-security-skills — Claude Code plugin and agentskills.io pack for securing LLM agents
- agentdojo-mcp — runs AgentDojo's benchmark against real MCP servers
- agentic-semgrep-rules — Semgrep rules for AI agent code
- ai-agent-incidents — incident database for AI agent failures
Other work: adaptive-knowledge-graph, aequitas, agentic-evolution-harness, HA-RohlikCZ.
Research, writing and photography at nawaar.pplx.app.