# Suraj Ranganath

Suraj Ranganath is an AI/ML engineer and UC San Diego Data Science graduate student working on AI agents, identity security, reinforcement learning, EEG tooling, graph machine learning, learning analytics, world-model memory, and LLM robustness.

## Primary Page

- [Homepage](https://surajr.dev/): One-page personal website with profile, selected work, timeline, GitHub activity, patents, publications, resume, and contact information.
- [Markdown profile](https://surajr.dev/index.md): Plain Markdown version of the website for language models, search agents, and retrieval systems.
- [RSS feed](https://surajr.dev/rss.xml): RSS 2.0 feed for site, selected work, research, and patent updates.
- [API catalog](https://surajr.dev/.well-known/api-catalog): RFC 9727 linkset catalog for public machine-readable endpoints.
- [Resume PDF](https://surajr.dev/Suraj_Ranganath_Resume.pdf): Current public resume.

## Stable Anchors

- [Contact](https://surajr.dev/#contact)
- [Selected work](https://surajr.dev/#selected-work)
- [BalkanID Copilot](https://surajr.dev/#balkanid)
- [StealthRL](https://surajr.dev/#stealthrl)
- [EEGPrep](https://surajr.dev/#eegprep)
- [PIPE-RDF](https://surajr.dev/#pipe-rdf)
- [Timeline](https://surajr.dev/#timeline)
- [Patents](https://surajr.dev/#patents)
- [Research and publications](https://surajr.dev/#research)

## Key Topics

- AI agents
- Identity security
- Knowledge graphs
- Graph machine learning
- LLM robustness
- AI-text detection
- Reinforcement learning
- EEG analysis
- Learning analytics
- World models
- Video diffusion
- Data systems

## Selected Work

- BalkanID Copilot: Agentic identity-security product over enterprise identity graphs using LLMs, knowledge graphs, text-to-Cypher, evaluation, monitoring, and governance workflows.
- Show, Don't TELL: Explainable AI-generated text detection with evidence spans.
- StealthRL: Reinforcement-learning red-team loop for AI-text detectors, with a public demo and GitHub repository.
- EEGPrep: Python port of EEGLAB, the popular toolbox widely used by EEG researchers.
- NAMO and NAMO-D in Marin: Open-source optimizer implementation for foundation-model training in Marin, mentored by Al Merose.
- KV Cache Quantization for Self-Forcing Autoregressive Video Generation: 33-method study for self-forcing autoregressive video generation.
- PIPE-RDF: LLM-assisted enterprise RDF benchmark generation.
- TrailKarma: Offline-first trail intelligence app with BLE relay and on-device ML.
- Multimodal Learning Analytics Dataset: Paper using EEG, eye tracking, GSR, video, and spatial reasoning data.
- VaultLens: Telegram personal knowledge-base agent.
- Where Bits Matter in World Model Planning: Mixed-bit world-model planning study.

## External Profiles

- GitHub: https://github.com/suraj-ranganath
- LinkedIn: https://www.linkedin.com/in/suraj-ranganath/
- Google Scholar: https://scholar.google.com/citations?user=5SOfSs8AAAAJ&hl=en
- Medium: https://medium.com/@surajranganath

## Contact

Professional inquiries: suranganath@ucsd.edu