2,385,015open jobs
148,810companies
289,953added this week
Browse all
Location
In office (Shanghai)
Seniority
Middle · 3+ years exp
Visa
Sponsors visas in Ireland; no register of this country names it

Confirmed on the employer's own hiring board on Nov 19, 2026. First seen by Alion on Nov 19, 2026. Propylon scores A on the Alion truth index.

Overview
Company
Impact
Profile match
Propylon is a legal and regulatory technology company, headquartered in Dublin, that builds software for drafting, publishing and researching legislation for parliaments and governments and for managing regulatory change in regulated companies, standards bodies and content publishers. With more than 25 years of work with laws, regulations and standards, it offers drafting, publishing, research and TrueSource products, Microsoft 365 integration and managed services, and has a US office in Chicago. Propylon is part of the RWS group, whose global careers board lists language, localization, AI data and technology roles worldwide.

Job Purpose

About the Role

You own two things: the software a person experiences from our humanoid, and the data that comes out of it.

The software is the speech pipeline that hears people in their language, the dialogue that answers from the venue’s knowledge, the skills that turn an answer into a greeting, a direction or a hand-off, and the operator console that lets one person keep the robot good on day one. You build these to the specifications of our HRI scientist, who designs the interaction and owns the content.

The data is every interaction the robot has and every demonstration we record with the robot and our capture devices. You run the pipeline that turns it into training-ready datasets: ingestion, quality control, annotation export and delivery in the formats robot-learning teams use, to the data specification our technical lead defines.

You are the upper layer of a small engineering team. A systems integration engineer owns the robot and exposes it to you as behavior primitives and sensor events; you do not decide what the robot should do, you make it do what the interaction designer specified, reliably and fast, and you make sure everything it does becomes data.

About Train AI

RWS TrainAI is the AI data and evaluation business of RWS Group, working with leading AI developers worldwide. We are building a physical AI practice: multilingual, culturally localized interaction for service robots, human evaluation of how robots behave around people, and the data pipelines behind both. Our advantage is a global, multilingual human network and a focus on behavior that is culturally appropriate, measured and auditable.

Job Overview

Key Responsibilities

Run the data pipeline

  • Own the server side of our robot data collection to the technical lead’s specification: ingestion from the robot and from capture devices, submission-time checks and batch quality control, episode segmentation and annotation export, and delivery in the formats clients use (LeRobotDataset, RLDS, MCAP).
  • Log every interaction (what the person said, what the robot heard, what it did, what the operator corrected) against the failure taxonomy the HRI scientist defines, with the consent and notice hooks our data team specifies, and report autonomy rate and operator calls per hour from day one.
  • Operate the observability and QC stack against live collection, keep data inside China where contracts require it, and keep the stack reproducible so a second robot in another location comes up from your documentation.

Build the voice and dialogue pipeline

  • Test the platform’s own speech stack in venue noise for turn latency and recognition accuracy per language, and build the alternative if it does not meet the bar: a LAN-hosted pipeline of streaming speech recognition, language identification and switching, an LLM dialogue layer grounded in the venue knowledge base, and multilingual speech synthesis, with the robot as a client and no internet dependency.
  • Keep the whole turn fast enough for a face-to-face conversation; measure and report latency per stage and hold it inside the budget.

Build the skill library and the knowledge base

  • Build the task-level skill library (greet, answer, direct, recommend, hand off) as behavior trees or state machines on the robot’s behavior primitives, to the HRI scientist’s specification of what the robot does and when, including the timing of speech, gesture and gaze.
  • Build the knowledge pipeline: the venue’s floor plans, points of interest, scripts and FAQs, and a tagged map of the venue, into a retrieval layer the dialogue answers from, maintained by the content owners without engineering help. Behavior and language are configuration, so a new venue or language is content, not code.

Build the operator console

  • Build the Wizard-of-Oz console to the HRI scientist’s workflow: live view of what the robot hears, sees and intends; approve, correct or trigger skills within the time a person will tolerate; runs on a laptop or tablet.

Skills & Experience

Required

  • 3+ years building production software, including data pipelines that ran against live data: ingestion, validation and quality control, format conversion, and the operations behind them.
  • Experience with multimodal or time-series sensor data (video, audio, IMU, robot joint states): synchronization, storage formats (Parquet, HDF5, MCAP or similar) and dataset versioning.
  • Hands-on experience with speech and language pipelines: streaming speech recognition, text-to-speech, and LLM integration with retrieval over a knowledge base, in at least one system that ran with real users.
  • Strong Python and one of TypeScript, C++ or Go; comfortable with Linux, Docker, GPU inference serving and local-network deployment.
  • Experience building operator or internal tools with a web front end, and integrating with a hardware API (ROS 2, a robot SDK or a device SDK) as a client.
  • Native-level Mandarin and professional working English; eligible to work in China.

Preferred

  • Voice-agent frameworks (Pipecat, LiveKit Agents, TEN or similar) and self-hosted open models (Qwen-class LLMs, Whisper-class speech recognition).
  • Multilingual language identification and switching, or conversational AI for a second language.
  • Robot behavior scripting with behavior trees or state machines, or real-time interactive systems such as games or animation.
  • Robot learning data conventions (LeRobot, RLDS) and an understanding of how imitation-learning or VLA models consume data; multimodal annotation or observability platforms.
  • Teleoperation or Wizard-of-Oz tooling and turning operator corrections into training data.
  • Robots or kiosks that interact with the public, and privacy and consent requirements for recorded interactions (PIPL, GDPR).
  • SLAM, navigation, or real-to-sim and simulation work. Venue maps, reconstructed scenes and simulated evaluation runs all flow through the data pipeline you own.

What success looks like

  • First 3 months: a live, multilingual interaction demo on the real robot in front of an audience. Not polished: scripted interactions plus open Q&A on a venue knowledge base, in two languages with automatic switching, with one operator on your console; every interaction logged and exportable for annotation from that first demo; speech-stack decision made from measured data.
  • By 6 months: capture-device and robot recordings ingested, QC’d and delivered in LeRobotDataset / RLDS / MCAP against the technical lead’s quality metrics; autonomy rate and operator calls per hour reported weekly; skill library and knowledge base maintained by the content owners without you.
  • By 12 months: observability and QC stack running against live collection without engineering intervention; a new language added as configuration in under a week; the stack deployed on a second robot in another location from your documentation.

Life at RWS

Life at RWS- If you like the idea of working with smart people who are passionate about growing the value of ideas, data and content by making sure organizations are understood, then you’ll love life at RWS.

Our purpose is to unlock global understanding. This means our work fundamentally recognizes the value of every language and culture. So, we celebrate difference, we are inclusive and believe that diversity makes us strong. We want every employee to grow as an individual and excel in their career.

In return, we expect all our people to live by the values that unite us: to partner, putting clients fist and winning together, to pioneer, innovating fearlessly and leading with vision and courage, to progress, aiming high and growing through actions and to deliver, owning the outcome and building trust with our colleagues and clients.

RWS embraces DEI and promotes equal opportunity, we are an Equal Opportunity Employer and prohibit discrimination and harassment of any kind. RWS is committed to the principle of equal employment opportunity for all employees and to providing employees with a work environment free of discrimination and harassment. All employment decisions at RWS are based on business needs, job requirements and individual qualifications, without regard to race, religion, nationality, ethnicity, sex, age, disability, or sexual orientation. RWS will not tolerate discrimination based on any of these characteristics.

RWS Values

Get the 3Ps right - Partner, Pioneer, Progress - and we´ll Deliver together as RWS.

Recruitment Agencies: RWS Holdings PLC does not accept agency resumes. Please do not forward any unsolicited resumes to any RWS employees. Any unsolicited resume received will be treated as the property of RWS and Terms & Conditions associated with the use of such resume will be considered null and void.

Free account
Stop reading job ads. Get the ones that fit.
One free account turns this page into a shortlist built around your stack, your level and your pay.
Match on every job. Stack, seniority, pay and location, scored against your profile.
2,385,015 open roles. Read straight off company career pages, refreshed every day.
Unlimited applications. Every one you send is tracked in one place, on-site or on a company board.
3 tailored CVs a month. Rewritten for the exact job you are applying to. Included free.
Create a free account Continue with Google
Free forever. No card. Under a minute.

Your match

How well do you fit this role?
Two answers are enough for a real match. No account needed.
Check my fit
Answers stay in this browser until you create an account.

Recommended for you based on this role

Data Science
Similar stack
Same company
Shanghai
In office · 10+ years exp · Bachelor's Degree · Guangzhou
Python
Java
SQL
Python
pySpark
Databases
PostgreSQL
Databricks
Delta Lake
MS SQL
Apache Kafka
AI/ML
Spark
Airflow
AI Agents
LLM
LLM Guardrails
Machine Learning
DevOps
GCP
Prometheus
Azure
CI/CD
Git
AWS
Kubernetes
Grafana
Amazon EKS
Azure AKS
Analytics
ETL/ELT
Management
Agile
Apply
In office · Beijing
Java
SQL
AI/ML
Flink
Apply
In office · Internship · Shanghai
Java
Kotlin
TypeScript
C++
AI/ML
Function Calling
VLM
Apply
In office · Beijing
Python
SQL
AI/ML
DeepSpeed
vLLM
SGLang
Pandas
Megatron-LM
DevOps
Prometheus
Grafana
Apply
In office · Beijing
Python
JavaScript
C++
Databases
Redis
ElasticSearch
Apache Kafka
AI/ML
Spark
VLM
Flink
LLM
Browser Agents
Apply
In office · Full-Time · 7+ years exp · Master's Degree · Shanghai
Apply
In office · Full-Time · 8+ years exp · Bachelor's Degree · Shanghai
Python
C++
AI/ML
Copilot
Claude Code
OpenAI Codex
DevOps
Azure DevOps
Azure
Git
GitHub
Linux
IoT
FreeRTOS
Life Sciences
Design of Experiments
Apply
In office · Full-Time · 8+ years exp · Bachelor's Degree · Shanghai
Chips/EDA
Altium Designer
Life Sciences
Design of Experiments
Apply
In office · Shanghai
DevOps
Linux
Apply
In office · Full-Time · Shanghai
Python
MATLAB
Apply
See all jobs
This is one of many
2,385,015 more open roles from verified company boards, updated every day.