I work on transformers, agents, and local-first AI, with a focus on applied ML, post-training, and deployment. My background spans protein interaction prediction, voice training for real study participants, molecular discovery, and robotics.
Currently: Based in Milpitas, CA · UC Computer Science, BS Honors, 3.9 GPA, August 2026 · 3 published papers
| Project | What it is |
|---|---|
| 🤖 sim2real-vla | Closed-loop VLA post-training, MuJoCo Franka, FSDP2 on 4x A100s |
| 🎓 ProjectSchola | Socratic AI tutor for Claude Code, verified citations, spaced review |
| 🧪 chemcrow-orchestrator | Local ChemCrow agent, Ollama-backed reasoning LLM |
| ⚛️ OPV-Chemist | RL framework for organic photovoltaic discovery |
| 👁️ QueryScope | Local visual Q and A, no cloud calls, no API keys |
Closed-loop vision-language-action post-training in a domain-randomized MuJoCo Franka tabletop sim. SigLIP plus Qwen2.5-0.5B, with discrete and flow-matching action heads, FSDP2 training on a single node with 4x A100s, and a decoupled GRPO actor-learner setup for low-latency inference.
| Stack | Focus |
|---|---|
| Python, PyTorch, MuJoCo, SigLIP, Qwen2.5-0.5B, FSDP2, DeepSpeed ZeRO-3 | VLA post-training, behavior cloning, GRPO-style RL, CUDA-graph inference |
A Socratic AI tutor for Claude Code. Learn any topic, or teach from your own PDFs, textbooks, docs and videos, with verified citations, quizzes, spaced review and live Obsidian notes. Local-first.
| Stack | Focus |
|---|---|
| Python, Claude Code, Obsidian | Socratic tutoring, verified citations, spaced repetition, local-first learning |
A ChemCrow fork with a local Ollama-backed reasoning LLM, a custom-tool scaffold, and a Streamlit UI for chemistry reasoning without sending work to a hosted model.
| Stack | Focus |
|---|---|
| Python, LangChain, Ollama, RDKit, Streamlit | Local LLM agents, chemistry tools, AI for science |
Extended CycleChemist into a hardened, DFT-validated RL framework for organic photovoltaic discovery, combining generative transformers, structural plausibility gating, and TD-DFTB oracle validation.
| Stack | Focus |
|---|---|
| Python, PyTorch, Transformers, RDKit, RL | Generative molecular design, RL for materials discovery, OPV |
Local visual question answering. Upload an image, ask a question, and get an answer from a vision-language model running entirely on your machine via Ollama. No cloud calls, no API keys.
| Stack | Focus |
|---|---|
| TypeScript, React, FastAPI, Python, Ollama, Qwen3-VL | Vision-language models, local-first AI, VQA |
| Paper | Venue |
|---|---|
| SPPIDER-seq: sequence-based partner-aware predictor of protein-protein interaction sites | Bioinformatics, 2026 |
| TruVox Web-Based Software for Vocal Pitch Training in Transgender Women: Development and Single-Session Evaluations | JMIR Formative Research, 2025 |
| Terminated Clinical Trial of TruVox: A Web-Based Intervention to Train Vocal Pitch and Resonance in Transfeminine People | Behavioral Medicine, 2026 |
AI/ML Engineer Intern · University of Cincinnati (OaSiS project fund) May 2026 - Present · Cincinnati, OH
Building an end-to-end AI pipeline for molecular discovery, with a transformer-based molecular language model fine-tuned with reinforcement learning, and ML infrastructure scaled for large molecule datasets.
AI/ML Engineer Intern · OPV Materials Discovery Jan 2026 - May 2026 · Cincinnati, OH
Built a multimodal generative model for donor-acceptor candidates for organic photovoltaic materials, with LoRA fine-tuning and a FAISS-based RAG pipeline for synthesis procedures and process constraints.
AI/ML Engineer Intern · Cincinnati Children's Hospital Sep 2024 - Dec 2024 · Cincinnati, OH
Contributed to SPPIDER-seq, a partner-aware cross-attention model using ESM-2 embeddings, reaching up to 0.797 AUROC and 0.269 MCC on its peptide-centric blind test. Credited roles: data curation, software, and writing review and editing.
AI/ML Engineer Intern · University of Cincinnati x NYU (collaborative project) Feb 2024 - Sep 2024 · Cincinnati, OH
Led a team of 4+ engineers and researchers to design and deploy TruVox, an AI-powered experimental platform for digital voice-training research using Python, PostgreSQL, and Redis, for large-scale participant studies.
| Recognition | Details |
|---|---|
| 🎓 Education | BS Honors, Computer Science, University of Cincinnati, 3.9 / 4.0, August 2026 |
| 📚 Publications | 3 published papers: SPPIDER-seq in Bioinformatics, plus two TruVox papers in JMIR Formative Research and Behavioral Medicine |
| 🔬 Research | SPPIDER-seq: partner-aware cross-attention with ESM-2 embeddings, up to 0.797 AUROC and 0.269 MCC on peptide-centric blind test |
| 👥 Leadership | Led 4+ engineers and researchers on TruVox, UC x NYU collaborative project |
| 🤖 Robotics / VLA | sim2real-vla: closed-loop VLA post-training, single-node 4x A100 FSDP2, GRPO actor-learner |
I build AI tools that ship, not just demos.