On Lounge since Oct 2026
James W. Niu
Member #1,269DeveloperL2Enterprise AI Architect, Basis Technologies
Open to Staff+ AI roles. Applied AI/ML, FDE. Multi-agent systems, LLM evals, retrieval, voice AI. LangGraph, MCP, A2A, RAG, MLOps. Measured
How karma works. Everyone starts with 5. You get +1 each time another member upvotes one of your posts or comments (your own upvotes don't count), and lose it if they take the upvote back. Post what you're building, answer questions and help other members to earn more.
Achievements
Plugged In ×3Connected sources: GitHub, Hugging Face, LinkedIn, X, website.How: Connected 3 sources (GitHub, LinkedIn, X or a website) in Settings
Plugged In×3Featured
Shipper ×2Projects on your profile.How: Added 5 projects to their profile
Shipper×2
Pinned projects
Autonomous AI filmmaking for ad creative, where the evals are the product. A cloned voice and a generated avatar, alpha matting, and monocular depth on Apple Silicon warped into a 77-view light field, gated by 13 probes and 16 of 16 thresholds derived from labelled pass and reject pairs so it runs with nobody watching.
3dad-techalpha-mattingcomputer-visiondepth-estimation
Multi-agent causal simulation of 1,000 AI-agent couples in Rust (Bevy ECS) and Python. Randomized controlled trials on agent memory with bootstrap CIs, permutation tests, placebo and counterfactual twins, and SHA-256 replay. A React and PixiJS 2D world, plus a Claude LLM that turns plain-language shocks into validated, deterministic inputs.
agent-based-modelingagent-memoryai-agentsbevycausal-inference
Know whether a change to your simulation actually caused the result, or was just luck. A reproducible randomized-controlled-trial harness for agent-based models with common-random-number twins, randomized arms, a byte-identical placebo, and fingerprint-verified replay, in Python, TypeScript and Rust with one identical fingerprint.
abmagent-based-modelingcausal-inferencecommon-random-numberscomputational-social-science
Real time drone AI for safe and adaptive flight that separates drift from hallucination. Multi agent reinforcement learning with Proximal Policy Optimization ("PPO") for path finding and coordination, served on a FastAPI backend with weight loading at startup, test isolation, and Docker deployment to bridge research into production
ai-safetyci-cd
dockerdrone
fastapi
An autonomous ad pipeline that shoots with no director, grades pixels and audio instead of text, and picks a video engine for each audience. Multi-agent Python, ffmpeg, OpenCV and MediaPipe probes, LLM-as-judge, a hand-labelled golden set and eval-gated CI. The evals do all three, and it spends nothing until they pass.
ad-techagent-evaluationagentic-aiai-agentsai-video
WhatsApp's official API is Business-only and never touches your personal chats or groups. This wraps the WhatsApp Web protocol in a local HTTP API any client can drive: agents (OpenClaw, Claude Code), production AI, cron, curl. Deterministic enough to trust with a real account: dry-run default, gated mutations, queryable action IDs.
agent-toolsagentic-aiai-agentsautomationcli
About
Open to Staff+ AI roles. Applied AI/ML, FDE. Multi-agent systems, LLM evals, retrieval, voice AI. LangGraph, MCP, A2A, RAG, MLOps. Measured before believed.
Experience
Sr. Staff AI Engineer | Multi-Agent Systems | Model/LLM Evaluation, Multimodal | Voice AI, Reinforcement Learning, RAG | LLMOps, MLOps, Python, Fine-Tuning | PyTorch/TensorFlow, Docker/Kubernetes | Columbia MSc | LSE BSc
Sr. Staff AI Engineer · Basis
Principal AI Solutions Architect · Advantage Solutions
AI Solutions Engineer · Leadership Circle
Data & AI Architect · Medlytix LLC
Forward Deployed Engineer (FDE) · GeoAlliance Consultants Pte Ltd
Machine Learning (ML) Engineer · M Capital Group (MCG)
Investment Banking Analyst · Nymex Capital Partners LLC