Open-source systems for AI alignment research, developer automation, and local-first agentic tooling.
- About
- Featured Projects
- Tech Stack
- GitHub Trophies
- Activity & Stats
- Star History
- Roadmap
- Philosophy
- Contributing
- Connect
- Contact
This organization hosts research and engineering projects built by Mohit Yadav, spanning LLM alignment research, GitHub automation, developer productivity tools, and local-first multi-agent systems.
Projects emphasize reproducibility, maintainability, honest documentation, strong engineering practices, and practical software that runs on accessible hardware whenever possible.
| Repository | Description | Stack | Stars | Forks |
|---|---|---|---|---|
| dpo-vs-rlhf-alignmet-study | Empirical comparison of DPO vs. RLHF alignment on GPT-2, trained on the 160,800-pair Anthropic HH-RLHF dataset. Full pipeline, results, and paper draft (targeting EMNLP/COLING) included. | Python, PyTorch, HuggingFace | ||
| hiero-bot-py | Production FastAPI GitHub App for maintainer automation: PR health scoring, reviewer recommendations, stale-issue management, and a live analytics dashboard. 76+ tests, deployed. | Python, FastAPI, SQLAlchemy, APScheduler | ||
| mats-compute-admin | Compute resource administration toolkit for research fellowship programs — scholar provisioning, budget tracking, API key lifecycle, Slurm/HPC monitoring, and IAM auditing. 41 passing tests. | Python (stdlib-first) | ||
| HydraNet / Loopcutter | Local-first multi-agent coding assistant focused on one problem: detecting when an agent loops on a broken fix, rolling back, and forcing a genuinely different approach. Runs on consumer GPUs, no cloud calls. | Python, LangGraph, Tree-sitter, Docker | ||
| IssueScout | Full-stack GitHub contribution discovery platform that analyzes repositories, identifies contributor-friendly issues, detects linked pull requests, and ranks opportunities using evidence-driven confidence scoring. | FastAPI, React, TypeScript, Python | ||
| ptet-web | Informational website providing PTET-related resources in a clean, user-friendly format. | HTML | ||
| E-commerce | Responsive e-commerce platform with product browsing and cart functionality. | JavaScript | ||
| makemore | Character-level language modeling exercises exploring autoregressive generation fundamentals. | Python | ||
| AI-Resume-Builder | Tool for generating and formatting resumes with AI assistance. | — |
Star/fork counts update live via shields.io — no manual maintenance needed.
Python · FastAPI · React · TypeScript · Vite · PyTorch · SQLAlchemy · Docker · PostgreSQL · GitHub Actions · Tree-sitter
- Reproducibility over hype — results are published with numbers, including failures where relevant.
- Consumer hardware first — projects are built and benchmarked on hardware anyone can access, not assumed cloud budgets.
- Honest scope — READMEs describe what a project actually does before what it aspires to.
- Fork the relevant repository
- Create a feature branch (
feature/your-feature-name) - Commit your changes with clear messages
- Open a Pull Request describing the change and motivation
Repository-specific contribution guidelines, where they exist, take precedence over this general note.
Rendered live from GitHub's contribution data — updates automatically, no manual edits required.
- Release IssueScout v1.0 with a production-ready React + FastAPI architecture
- Ship
hiero-bot-pyv2 with live dashboard and PR health scoring - Publish DPO vs. RLHF empirical results and paper draft
- Build
mats-compute-adminwith full test coverage - Take
Loopcutter(HydraNet) from single-machine loop to multi-node coordination - Run Loopcutter against a real slice of SWE-bench Lite
- Submit the DPO vs. RLHF paper to EMNLP / COLING
- Expand
hiero-bot-pyreviewer recommendation with embedding-based file similarity
Reach out via GitHub Issues on the relevant repository, or anthropicbots@gmail.com.
© 2026 AnthropicBots · Maintained by Mohit Yadav · Individual repositories are licensed under MIT unless otherwise noted.