Work
AI systems and infrastructure: retrieval, evaluation, agents, and data pipelines.
Experience
Data Engineer Intern
TikTok, San Jose
Enterprise retrieval for internal AI agents; livestream revenue-diagnosis copilot.
Details Close
- — Designed a graph-based enterprise retrieval skill over Hive tables, datasets, dashboards, documents, and millions of metadata fields and values, using query decomposition, contextual rewriting, graph expansion, and reciprocal rank fusion (RRF).
- — Built a benchmark from on-call Q&A pairs to evaluate agent results. The skill was integrated and published as an internal skill.
- — Built a livestream revenue-diagnosis copilot: LightGBM percentile prediction, comparison cohorts, and business rules flag revenue drops and abnormal metrics; retrieved business documentation grounds the recommendations.
- — Validated with operations stakeholders and integrated into their workflow.
Research Assistant — NLP lead
Sapling Lab, CMU
Theory-of-Mind attribution toward AI in long-term interaction.
Details Close
- — Lead NLP research on an NSF- and OpenAI-funded interdisciplinary project.
- — Developing a multi-turn user simulation that varies user traits, assistant social behavior, memory, and scenario.
- — Developing turn-level LLM judges and heuristic measures to evaluate how attributions change across a conversation. Manuscript in preparation.
Research Assistant
Chimps Lab, CMU
LLM red-teaming and image-generation bias auditing.
Details Close
- — PersonaTeaming (UIST 2026, with Apple Research): designed the persona-generation algorithm for adaptive LLM red-teaming.
- — Evaluated 11 experimental conditions on 6 target LLMs. Improved mean attack success rate by 46% and attack diversity by 7% over the RainbowPlus baseline.
- — Built and deployed the PersonaTeaming Playground on AWS, used in an industry practitioner study.
- — Built bias-auditing infrastructure for image generation: persistent history, caching, and batching at 100 images per prompt, with interactive visualizations.
Full Stack Engineer & Tech Lead
Enterview (AI startup), Pittsburgh, PA
AI interview-preparation platform.
Details Close
- — Led a team of four engineers; owned technical product decisions.
- — Designed user-context data models and backend APIs, integrating an LLM and speech recognition to evaluate spoken answers.
- — Reduced feedback latency by optimizing WebSocket handling.
Also
- Software Engineer Intern · AlpSoft, Singapore. Enterprise sites for EnterpriseSG; penetration-test fixes and AWS deployments. May 2025 – August 2025
- Research Intern · Tsinghua University, THUNLP — OpenBMB. REST APIs as LLM-callable tools; embedding-based clustering for tool retrieval. May 2023 – August 2023
- Teaching Assistant · CMU — Great Ideas in Theoretical Computer Science (15-251). Proof and theory tutoring; study guides and exam reviews. February – May 2025
- IS Consulting Project · ProtectPT (Non-Profit), Pittsburgh, PA. Searchable database for a local anti-fracking non-profit. Spring 2024
Projects
standin
October 2026 – PresentPersonal agent that answers as you and discloses only approved information.
Details Close
- —Local store of sourced, bitemporal memories, including explicit unknowns.
- —Default-deny airlock: only approved, versioned, reversible bundles reach the public agent.
- —Hybrid retrieval (keyword search and local embeddings, fused with RRF); eval harness that fails on any leak.
Golfie
January 2026 – PresentGolf swing analysis from phone video: fault detection, drills, and coach review.
Details Close
- —Three-phase pipeline: pose estimation, rule-based fault detection across five swing phases, LLM-generated drill plan.
- —Per-video pose tiers, from RTMPose on CPU to GPU models on Modal.
- —Coarse pass to locate the swing, then a dense pass at native frame rate to capture impact.
Daybook
October 2026 – PresentLocal-only weekly review of Mac Screen Time.
Dealer’s Choice
June 2025 – PresentOnline multiplayer platform for home-game poker variants.
Details Close
- —Custom variants with real-time multiplayer over WebSocket.
- —ECS in private subnets behind a public ALB; CI/CD for testing and deployment.
- —Variant rules: Personal.
Brawl Lab
October 2026 – PresentBrawl Stars final-pick recommender. Calibrated logistic regression trained on about 400,000 Ranked sets.
Table Forty Nine
October 2026 – Present1v1 online poker card battler. In design.
Publications
PersonaTeaming: Persona-Conditioned Automated Red Teaming
UIST 2026 · CMU–Apple collaboration
Persona-conditioned automated red-teaming.
Details Close
Conditions adversarial prompt mutation on personas, including a dynamic method that generates and scores candidate personas. Designed the persona-generation algorithm, implemented the workflow, and ran the evaluation: 11 conditions, 6 target LLMs. Over RainbowPlus: +46% mean attack success rate, +7% attack diversity.
Solver-Guided Fine-Tuning for Accurate and Explainable Poker Decisions
10-423/623 Generative AI, Spring 2026 · with Cassandra Shi & Harry Huang
Solver-guided post-training of an 8B model for accurate, explainable poker decisions.
Details Close
Llama 3 8B trained on 3,000 SFT examples and 2,534 DPO pairs from PokerBench solver labels and verifier-ranked Llama 3.3 70B explanations; rule-based verifier validated against 200 blind human judgments. LoRA SFT: solver-action match 18.0% → 56.2%, verifier score 0.426 → 0.645. DPO improved explanations at a small cost to action accuracy.
Read the paper (PDF)Mutual Theory of Mind in Multi-Turn Human–AI Conversation
Manuscript in preparation
Turn-by-turn mind attribution in multi-turn human–AI conversation.
Details Close
Multi-turn user simulation varying user traits, assistant social behavior, memory, and scenario; measured with turn-level LLM judges and heuristic measures. Sapling Lab.
Education
MS in Language Technologies (Artificial Intelligence)
Expected August 2027Carnegie Mellon University, Pittsburgh, PA
Current coursework Close
Search Engines (11-642), AI Agents (11-768), Multimodal Machine Learning (11-777), Introduction to Research in Language Technologies (11-705), Directed Research (11-910)
BS in Computer Science and Information Systems
May 2026Carnegie Mellon University, Pittsburgh, PA
Minor in Logic and Computation
GPA: 3.86/4.0
Selected coursework Close
Machine Learning (10-301), Generative AI (10-423), Natural Language Processing (11-411), Algorithm Design and Analysis (15-451), Distributed Systems (15-440), Introduction to Computer Systems (15-213), Parallel and Sequential Algorithms (15-210), Bug Catching: Automated Program Verification (15-414), Foundations of Software Engineering (17-313), Database Design and Development (67-262), Probability and Statistical Inference I–II (36-235/236), Game Theory (80-305)
Skills
- Programming
- Python · SQL · C · JavaScript · TypeScript · Data structures · Database design · Distributed systems
- Machine learning
- PyTorch · vLLM · Hugging Face · TRL · PEFT / LoRA · SFT · DPO · Transformers · Multimodal ML
- Agents
- Agentic RAG · Agentic search · Tool use · LLM evaluation · Skills · MCP · Context management
- Data
- PostgreSQL · SQLite · Feature engineering · Data pipelines · Information retrieval · Heuristic search
- Backend & infrastructure
- FastAPI · NestJS · REST APIs · WebSocket · Docker · AWS (EC2, ECS) · Terraform · CI/CD