PULSE — Probing Underlying Latent Safety Embeddings in Robot Foundation Policies
[ active ]Latent failure prediction inside frozen VLA policies (OpenVLA, ACT). Submitting to CoRL. With Dr. Wei, Da Vinci Lab @ FAU.
hello —
Anik Sahai · AI researcher
This site is less of a portfolio and more of a public workspace — ongoing research, unfinished ideas, things I'm building, and questions I haven't answered yet.
— Anik
/now
Updated May 2026. Inspired by the
nownownow.com convention.
the thread
The phrasing keeps changing. These are the four sentences that have survived the longest.
/projects
Three currently running. Full ledger → /projects
Latent failure prediction inside frozen VLA policies (OpenVLA, ACT). Submitting to CoRL. With Dr. Wei, Da Vinci Lab @ FAU.
Rebuilding a 2018 GEM e4 into a self-driving vehicle from the control stack up. With FAU.
A benchmark for evaluating physical reasoning in LLMs through sustained interaction with a live MuJoCo simulator. Submitted.
/notes
Short writing. Some of it is wrong; that's fine — it's a notebook.
AuthAI is the most fun thing I've built in a long time — partly because I'm building it with my dad, partly because we're testing it with real doctors and providers instead of imagining what they want.
Just finished recording the first EthicsLabs lesson. Turns out making the content is the easy part — getting it in front of kids on social is way harder than I thought.
Feels good seeing it slowly become a real project instead of an old cart sitting in the lab. Steering and braking integration is going to hurt.
all notes →/work
Companies I've founded, labs I work in, and places I've shown up to learn the shape of a real workflow.
Self-employed · Hybrid · Boca Raton, FL
Founded AuthAI after seeing how often patients get stuck waiting — not because the care is unclear, but because the process is. AuthAI streamlines prior authorization by capturing the clinical story accurately and turning it into what payers need, fast. Building it with my dad and testing it directly with doctors and provider offices.
Nonprofit · Self-employed · On-site · Boca Raton, FL
Leading an initiative dedicated to bridging the gap between Gen Z and the rapidly evolving world of AI. I oversee program design, partnerships, and strategic direction so the work highlights real ethical questions, real-world applications, and emerging opportunities in AI. At its core: make AI more understandable, accessible, and human-centered, and amplify the voices of the people who will inherit the technology.
Part-time · Hybrid · Hybrid
Working alongside Dr. Hahn on AI, robotics, and embodied intelligence research — with a focus on AI alignment and safety: how autonomous systems make decisions, interact with humans, and operate reliably in the real world. I assist with embodied AI experiments, analyze emerging AI technologies, and translate technical work into practical applications. I also mentor students working on AI and robotics projects, and help communicate MPCR's mission to external supporters.
Part-time · Hybrid · Boca Raton, FL
Built and deployed custom AI models — recommendation engines, predictive forecasting, NLP chatbots, automated QC systems — and the data pipelines around them. Worked mostly in Python, PyTorch, TensorFlow, and scikit-learn on AWS / GCP, covering data engineering, model optimization, and MLOps so launched systems stayed solid, interpretable, and improvable through feedback and retraining.
Apprenticeship · Remote · Boca Raton, FL
Conducted market research on AI-driven revenue cycle management — automated coding, claims processing, denial prediction — mapping established players and emerging disruptors. Evaluated technical capabilities, regulatory positioning, and growth potential, and wrote investment theses on companies across the healthcare AI / RCM space.
Apprenticeship · On-site · Boca Raton, FL
Coordinated patient care through multi-step documentation, intake, and scheduling — and supported physician research with data collection and clinical documentation. Served as a primary point of contact for patient inquiries, which is where most of what I now know about clinician/patient workflow friction actually came from.