代理系统工程师实习生（2026年夏季）
Agentic Systems Engineer Intern (Summer 2026)
我们的使命
在Phygtl，我们正在解决学生社交孤立的问题——这一问题影响着超过60%的当今学生，导致抑郁和焦虑情绪上升。
我们构建的技术可以将孤立感降低40%，通过帮助学生通过共享的现实时刻建立真实的校园联系，连接物理世界和数字世界。
我们是谁
总部位于旧金山，我们的团队包括来自斯坦福、加州大学伯克利分校、麻省理工学院和卡内基梅隆大学的博士，以及来自Niantic、Meta、Magic Leap、Riot Games、Ubisoft和Zynga的开发者。由一位硅谷连续创业者领导，我们因拯救下一代免于社交退化这一使命而团结在一起。
我们的护城河
Phygtl构建系统，帮助学生在现实中见面并建立身份。Vyry是一个上下文感知的引导引擎，任务最终以物理行动结束。留存率通过参与获得——而不是通知。
职位描述
- 你将加入开发Vyry的团队，这是一款移动应用，通过简短、协调的校园任务帮助学生在现实中见面。
- 目前，我们的重点是任务参与度优化：在用户通过聊天、语音和AR Meet协调时，保持同龄人停留在任务界面，避免状态错误、掉线或实时行为中断。
- 这个职位不是关于“构建聊天机器人”。
- 而是关于在实时社交产品中交付实时智能，其中状态正确性、延迟和用户信任至关重要。
- 你将负责的工作
- 设计上下文感知的任务逻辑（物理+社交+时间信号）
- 检测协调过程中的停滞或掉线情况
- 提出能提高完成概率的结构化微操作建议
- 构建与可衡量行为结果相关的评估循环
- 发布面向生产环境的组件
你将在现有的实时系统之上工作——而不是重建基础设施。
我们寻找的人
- 你是计算机科学或工程专业的优秀学生
- 有实际系统发布经验（需提供GitHub）
- 能够处理状态转换和边缘情况
- 对AI系统充满好奇，不只是简单的提示使用
- 能够快速从想法到工作原型
你需要能够：
- 设计一个简单的状态感知流程
- 思考失败模式
- 解释权衡（延迟、成本、可靠性）
- 将技术决策与可衡量的结果联系起来
如果你有以下经验会更佳：
- 自主工作流
- 多人或事件驱动系统
- 评估流水线
查看英文原文
Our Mission
At Phygtl, we are solving the problem of student social isolation — a challenge affecting 60%+ of today’s students, contributing to rising depression and anxiety.
We build technology that can reduce isolation by up to 40%, by helping students form real campus connections through shared real-world moments — bridging the physical and digital worlds.
Who We Are
Headquartered in San Francisco, our team includes PhDs from Stanford, UC Berkeley, MIT, and Carnegie Mellon, alongside builders from Niantic, Meta, Magic Leap, Riot Games, Ubisoft, and Zynga. Led by a Silicon Valley-based serial entrepreneur, we’re united by a mission to rescue next-gen from Social Atrophy.
Our Moat
Phygtl builds systems that help students meet and build identity in real life. Vyry is a context-aware initiation engine where quests end in physical action. Retention is earned through participation — not notifications.
The Role
- You’ll join the team building Vyry, a mobile app that helps students meet in real life through short, coordinated campus quests.
- Right now, our focus is Quest Participation optimization: keeping peers on the quest screen while they coordinate via chat, voice, and AR Meet, without state glitches, drop-offs, or broken realtime behavior.
- This role is not about “building chatbots.”
- It’s about shipping realtime intelligence inside a live social product, where state correctness, latency, and user trust matter.
What You’ll Work On
- Designing context-aware quest logic (physical + social + temporal signals)
- Detecting stall or drop-off conditions during coordination
- Suggesting structured micro-actions that increase completion probability
- Instrumenting evaluation loops tied to measurable behavioral outcomes
- Shipping production-facing components
You will operate on top of existing real-time systems — not rebuild infrastructure.
What We’re Looking For
You are:
- A strong computer science or engineering student
- Someone who has shipped real systems (GitHub required)
- Comfortable reasoning about state transitions and edge cases
- Curious about AI systems beyond simple prompt usage
- Able to move from idea → working prototype quickly
You should be able to:
- Design a simple state-aware workflow
- Think through failure modes
- Explain tradeoffs (latency, cost, reliability)
- Tie technical decisions to measurable outcomes
Bonus if you’ve built:
- Agentic workflows
- Multiplayer or event-driven systems
- Evaluation pipelines for AI systems
- Side projects used by real users
How We Evaluate
We prioritize:
- Systems clarity (state + failure thinking)
- Execution speed
- Ownership
- Depth over pedigree
Why Join Phygtl
Work on a Real Behavioral Problem. You won’t optimize clicks. You’ll design systems that influence real-world human coordination and identity formation.
This is AI applied to physical behavior — not interface optimization.
Ship Systems That Matter
Your work will tie directly to:
- IRL Quest Completion Rate
- D30 Retention
You won’t build experiments that disappear. You’ll ship components used by live campus communities.
Learn From Operators Who’ve Built at Scale
You’ll work with the foundering team, engineers and researchers from:
- Founder DNA
- Roblox (LLM + search systems)
- Niantic (AR + real-world coordination)
- Stanford / Berkeley / MIT / CMU
Expect depth, not hand-holding.
Operate in a High-Bar, Low-Bureaucracy Environment
We value:
- Meritocracy over pedigree
- Shipped systems over slide decks
- Clear thinking over buzzwords
Ownership comes early. Responsibility scales with performance.
Build a Rare Skill Set
You’ll gain experience at the intersection of:
- Agentic AI systems
- Real-time coordination
- Behavioral metrics
- Spatial computing
Few internships operate at this boundary.
Compensation & Structure
- Competitive compensation
- Flexible remote structure
- FT 3 months or PT 6 months
- High ownership, small team
Apply
- CV
- GitHub
- A concise architecture walkthrough of one system you built
- One failure you debugged (what broke, why, what changed)
- A short note (≤300 words) on how you’d increase IRL Quest Completion by 15%
We review for depth, not buzzwords.
Originally posted on Himalayas