📍 New York | 🔬 AI Safety & Agents Researcher | 🛠️ Open-Source Builder
I build open tools for researchers and study how language models behave when the stakes are real.
- 🔭 ResearchScope (110+ ⭐) — open research intelligence for CS papers: what matters, who drives it, what to read first
- ⚡ cli-faq-shortcuts (116+ ⭐) — turn repeated prompts into slash commands, mined from your own Claude Code history
- 🌙 nightaudit (30+ ⭐) — your AI works the night shift: read-only reviews, one digest every morning
- 🧭 Agent Command Atlas (24+ ⭐) — searchable atlas of commands across 6 AI coding agents
- 🎯 SkillsAllYouNeed (25+ ⭐) — directory of every first-party skill across 13 AI ecosystems
- 🔭 ResearchScope — open research intelligence for CS papers, with data on 🤗 researchscope-papers
- 🔍 CiteLens — finds the citing papers that matter most, ranked by impact and influence
- 🇧🇩 BanglaNLP Hub — community catalog of Bangla NLP papers, datasets, models and benchmarks
- 🎓 GradTracker — community tracker for Fall 2027 PhD/MS applicants: professor openings and fee-waiver webinars
- ⚡ cli-faq-shortcuts — turn repeated Claude Code, Codex and Cursor prompts into reusable slash commands
- 🌙 nightaudit — read-only reviews of your projects while you sleep, one digest every morning
- 🧭 Agent Command Atlas — searchable atlas of commands, flags and shortcuts across Claude Code, Codex, Gemini CLI, Cursor, Copilot CLI and Muse Code
- 🎯 SkillsAllYouNeed — every first-party skill and tool across 13 AI ecosystems
- 📐 promptlean — token-efficient prompt library: 120 prompts, 3 variants each (Lean / Balanced / Max Quality)
- 💥 blast-radius — agent skill: classify every affected row before a bulk write touches any of them
- 🔁 promptalias — reusable prompts as Agent Skills for Claude Code, Codex and Cursor
- 🧰 ECC — agent harness performance optimization: skills, instincts, memory and security
- 🔧 agent-repair — when a multi-step LLM agent fails, where should recovery begin: repair or restart?
- 🗣️ alexa-language-bridge — Alexa+ MCP add-on for non-English household members
- 📰 bangla-multimodal-political-stance — can the headline + photo predict Bangla news political leaning?
- 📜 banglachhanda — rule-based Bangla meter (chhanda) scanning: guidelines, baseline, pilot corpus
- 🧠 adaptive-ai-scaffold — AI writing tools that scaffold student reasoning instead of answering for them
| Dataset | What's inside | License |
|---|---|---|
| researchscope-papers | 100,000+ CS papers, scored and sectioned | CC BY 4.0 |
| bangla-nlp-catalog | 712 papers, 63 datasets, 20 models, 9 tools | MIT |
| llm-skills-registry | 226 first-party AI skills across 13 ecosystems | CC BY 4.0 |
| promptlean-prompts | 120 prompts × 3 token-efficiency variants | MIT |
| banglachhanda | 153 Bangla poems (7,009 lines), scansion pilot | CC BY-SA 4.0 |
Also on 🤗 Spaces: ResearchScope · CiteLens · BanglaNLP Hub · Agent Command Atlas · SkillsAllYouNeed · promptlean
Refreshed daily by a GitHub Action. Languages exclude HTML, CSS, notebooks and TeX.
- Researching AI safety — how LLM agents fail, recover, and misbehave in multi-step settings
- Building open research infra — tools that help researchers find what matters without a paywall
- Bridging Bangla NLP — datasets, benchmarks and tools for a language with 300M+ speakers and limited resources
- Teaching agents to be useful — skills, commands and workflows that make AI coding agents practical
- 80+ publications · 900+ citations · h-index 14 on Google Scholar
- Open datasets powering research on Hugging Face
- Tools used across Claude Code, Codex, Cursor and Gemini CLI ecosystems
"Build the tool you wish existed, then give it away." — I make open research and agent tools because the best work happens when everyone has access.
Random Facts
- 80+ papers and still shipping code
- Run multiple Claude instances concurrently (it's a lifestyle)
- Bangla is my first language — building NLP for it is personal
- Powered by late-night coding sessions and strong coffee