Building interests into projects you can read.
I’m Kevin Li — MSCS at NYU Courant. Trading tooling, LangChain wallet agents, and zk-SNARK messaging — all shipped end-to-end. Open to backend internships in the US.
Three systems, end‑to‑end.
Compact case studies of the backend, cryptographic, and product-facing work I’ve shipped between 2024 and 2026. Click through for the architecture and trade-offs.
Two routes, same risk model.
Crypto AI Agent is a FastAPI backend with two paths to the same
wallet-risk analysis. /analyze-wallet is a hard-coded Python pipeline —
feature extraction, rule engine, optional OpenAI explanation, Pydantic JSON out — and
runs without any API key. /agent/analyze-wallet wraps the same primitives
as three LangChain StructuredTools and lets the model decide which to
invoke from a natural-language question. Single-turn, no LangGraph; tool selection
delegated, not hard-coded. Tests cover both routes with stubbed agents.
What I’m on.
Concrete read on where my attention sits — what I’m building, what I’m practicing, and what I’m looking for next.
Crypto_AI_Agent FastAPI service — wiring the LangChain tool-calling agent to a real chain adapter (replacing mocks) and tightening the Pydantic contract between the deterministic and agent paths.
messageHash so they can’t be replayed.
Clear system boundaries first.
The intelligent layer comes second.
Every backend I build starts with the boring contract — input shape, validation, deterministic logic, fallback path, and the JSON it returns when everything goes right. The intelligent layer only earns its place after the baseline is honest.
That order shows up in every project on this site. The wallet agent is a deterministic feature extractor with optional explanation. The zk-messaging system separates proof generation from on-chain verification. The DApp keeps wallet auth, contracts, and the UI as three things that can be debugged independently.
I write code I want to read again in six months — one paragraph of clear types over a paragraph of clever ones.
Experience & education.
Two CS degrees, two internships in finance and database engineering, and one award worth keeping around.
Experience
2022 → 2023Reproduced and helped resolve system issues through frontend testing, built a transaction-data analysis model to support business decisions, and rewrote the testing workflow to catch issues earlier.
Optimized MySQL + Java workflows, designed query and indexing strategies, and built graph-style schemas for the company’s core data so relationships were retrievable in one hop.
Led scenes, dialogue, gameplay, and post-production for a small team. Shipped a polished build inside the two-day window.
Outstanding Design AwardEducation
2021 → NowGraduate coursework focused on systems, algorithms, machine learning, applied cryptography, and NLP. Building toward backend / AI infrastructure roles.
Coursework: Operating Systems, Algorithms, Machine Learning, Cryptography, NLP, Computer Graphics.
Dean’s List 2022–23What I reach for.
Less a list of badges, more the categories my projects actually live in.
Typed Python and TypeScript services with deterministic cores; the LLM layer is opt-in.
Building circuits, proving pipelines, and on-chain verifiers — keeping math out of the UI.
The systems and product muscles below the framework layer — what I actually got graded on.
Reach me at yl9314@nyu.edu — happy to talk about backend, AI infra, or fintech roles.