Skip to content
View LUOaini1213's full-sized avatar

Block or report LUOaini1213

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
LUOaini1213/README.md

Wenjie Luo · 罗文杰

Agentic AI product and evaluation, from a transport-engineering background. NUS MSc Civil Engineering (Transport), graduating Jan 2027 · Singapore.

Hard numbers stay in deterministic tools. The model only drafts. High-risk actions need a person. Whether the model was put in the right place is judged on a closed loop, not on the loss.

Site · English · CVs · e1576499@u.nus.edu · wenjiluo7@gmail.com

Dated write-ups: PROJECTS.md · commands and dependencies: RUNNABLE_PROJECTS.md

Three engineering cases: manufacturing outage → comparison → confirmed continuation, IFC type/unit → enrichment → independent readback, and driving-scene reconstruction → browser exploration. Each case links its demonstration, implementation and limits. 中文版.

Selected

Project What to inspect Demo or evidence
Glass Box / Track 1 Agent span waterfalls, redaction and durable OTLP delivery with explicit failure states. Official starter plus my observability layer; single-user PoC. No-key replay demo · Recovery and overhead record
fp16x3 Transformer Compensated FP16 GEMMs with FP32 input/output and precision checks. Recorded T4 performance gates include retained failures. Measured results and limits · Run and audit
LLM judge experiments Frozen judge replies, provenance auditing and question-cluster uncertainty on a personally constructed evaluation set. Sensitivity report
Manufacturing Planning Agent Compare from one snapshot, preserve completed work and WIP, then explicitly confirm before advancing. HiGHS solves; SimPy executes. Synthetic example, no factory dispatch. Recorded public replay · Case and actual local solver
NAVSIM ability ladder Full-split ablation of what caps open-loop planning, with paired bootstrap intervals. Results and reproduction
Counterask WebMCP tools ask for missing information instead of guessing. Browser interaction needs WebMCP support. Live store · Run locally

More engineering work: Civil Buddy · vLLM sm_75 throughput · EDA Copilot · CE5203 AYE weaving.

Choose a role and open matching project evidence · October 2026 follow-ups.

Stack

Python · PyTorch · CUDA / Triton · vLLM · FastAPI · TypeScript · Rust · SUMO · WebMCP

Popular repositories Loading

  1. ai-engineering-interview-questions-company-wise-zh ai-engineering-interview-questions-company-wise-zh Public

    AI 工程面试题(按公司分类)中文翻译版 | 原文: pallavi-shekhar/ai-engineering-interview-questions-company-wise (Apache-2.0)

    Markdown 2 1

  2. LUOaini1213.github.io LUOaini1213.github.io Public

    Personal site: agentic AI product and evaluation, transport engineering

    HTML

  3. civil-buddy civil-buddy Public template

    Civil Buddy — Agentic AI workspace for engineering. NL → tools → HITL → eval. Packing engine included.

    Python 1

  4. ce5203-aye-weaving ce5203-aye-weaving Public

    NUS CE5203: AYE westbound weaving bottleneck — YOLOv11 counts + SUMO control (VSL/RM), Group 12

    Python

  5. ce5212-llm-coordinator ce5212-llm-coordinator Public

    CE5212: LLM/rule traffic coordinator with semantic intent grounding. Course extract, not a vehicle planner.

    Python

  6. civil-buddy-workbench civil-buddy-workbench Public archive

    Archive: workbench merged into LUOaini1213/civil-buddy

    Rust