AI Solution Architect
About the job
- At least 10 years of experience as a Solution Architect, including at least 2 years of hands-on experience specifically in AI Architecture or AI Platform design.
- Fluent Chinese and English communication for high-level architectural discussions with global and local technical teams.
- Deep mastery of Kubernetes and full-lifecycle cluster management.
- Extensive experience designing and building MLOps and LLMOps platforms, including familiarity with Kubeflow, MLflow, and Databricks.
- Deep knowledge of core machine learning components, including Feature Stores, Model Asset Management, Supervised Fine-Tuning, and Model Serving.
- Familiarity with large-model fine-tuning and inference acceleration frameworks such as DeepSpeed, vLLM, SGLang, and TorchTune.
- High proficiency in Retrieval-Augmented Generation architecture design.
- Hands-on experience developing agentic applications with frameworks such as LangChain and LangGraph.
- Familiarity with vector databases and graph databases.
- Experience in high-availability system design, disaster recovery, role-based access control, and integrating enterprise components such as single sign-on, LDAP, and bastion hosts.
- Ability to ensure architectural designs comply with security standards and FinOps governance requirements.
- In-depth understanding of cloud-native architectures, including AWS and Alibaba Cloud.
- High proficiency in mainstream CNCF open-source technologies, particularly observability tools such as Prometheus, Grafana, OpenTelemetry, and ELK.
- Hands-on experience using Terraform, Ansible, Packer, and CloudFormation for automated deployment and system security hardening.
- Mastery of at least one backend development language, such as Python or Go.
- Experience building end-to-end pipelines with tools such as GitLab CI/CD and JFrog.
- Familiarity with AI hardware, including GPUs and high-performance CPUs, and storage technologies such as NAS and SAN.
- Ability to conduct performance benchmarking and select hardware for high-compute clusters.
- Understanding of data-center network architectures, including planning and configuring out-of-band networks and high-performance RDMA passthrough environments.
- Mastery of Linux operating systems, including RHEL or CentOS, and proficiency configuring the NVIDIA driver full stack and supercomputing driver toolchains.
- High proficiency in Scrum and Agile development processes and tools.
- Track record of leading large-scale AI platforms or projects and mentoring cross-functional teams to successful delivery.
- Lead deployment, administration, and full-lifecycle management of Kubernetes clusters, including node scaling, upgrades and rollbacks, certificate rotation, role-based access control administration, and cluster teardowns; develop cluster-operations automation and drive federated cluster management.
- Architect, build, and manage MLOps platforms using open-source or commercial frameworks; oversee the machine-learning lifecycle by integrating WebIDE, Feature Store, Model Registry, AutoML, pipelines, and model serving, and implement CMCT and drift-detection metrics.
- Design and develop platforms supporting the large language model lifecycle; integrate data assets, model cards, supervised fine-tuning, and OpenAI-compatible inference, and operationalize frameworks including DeepSpeed, TorchTune, vLLM, and SGLang.
- Design and implement enterprise AI gateways and integration platforms; establish FinOps, guardrails, and MAAS PrivateLink governance, and integrate with GCP Vertex AI, AWS Bedrock, Azure OpenAI, and Ali DashScope.
- Architect, design, and develop scalable AI and agentic applications using frameworks such as LangChain and LangGraph.
- Integrate applications across Kubernetes, Slurm, MLOps, LLMOps, AI Gateway, and data ecosystems including S3, relational databases, vector databases, and graph databases.
- Drive Domain-Driven Design to decompose APIs for complex business requirements and define service boundaries, contracts, and data models.
- Organize, lead, and mentor engineering teams delivering APIs, WebUIs, SDKs, CLIs, and applications.
- Establish and enforce coding standards, unit and integration testing protocols, and code-review processes.
- Design and optimize backend systems using REST, gRPC, GraphQL, SQL, NoSQL, Kafka, RabbitMQ, and caching mechanisms.
- Evaluate and implement architectural patterns including event-driven architecture, CQRS, Domain-Driven Design, and serverless, conducting trade-off analyses based on project needs.
- Oversee the full lifecycle of AI software development and ensure successful implementation and high-quality product delivery.
- Harden platforms and engineer reusable components using CNCF projects including Prometheus, Grafana, OpenTelemetry, Ray, ELK, and Kubeflow.
- Design and integrate observability solutions for metrics, logs, tracing, and alerting to monitor application performance, usage, and FinOps costs.
- Ensure architectures meet enterprise standards for high availability, disaster recovery and rollback, security compliance, and component reusability.
- Integrate systems with enterprise IT components, including single sign-on, LDAP, Git repositories, artifact repositories, cloud landing zones, Cloudflare, and AI gateways.
- Evaluate, benchmark, and select AI infrastructure hardware and storage solutions, and coordinate data-center power distribution and air or liquid cooling requirements.
- Architect network topologies and subnets; manage out-of-band, business, and high-performance RDMA passthrough networks using InfiniBand or RoCEv2; deploy centralized automated monitoring for network equipment.
- Administer Linux distributions and design automated security-hardening protocols for large server clusters; manage supercomputing drivers and toolchains including Ansible, Packer, NVIDIA and RDMA stacks, storage drivers, and containerization.
- Manage private-cloud environments, including IAM, data-center direct-connect, cloud-service security, and data encryption; drive Infrastructure as Code automation using Terraform, CloudFormation, and CLI tools.
- Configure and maintain enterprise bastion hosts and oversee super-user access controls.
- Drive large-scale, role-based project execution using standard Project Management Methodologies and align delivery with pharmaceutical-industry compliance standards.
- Champion Agile and Scrum; perform work-breakdown-structure decomposition, manage Jira backlogs, and coordinate cross-team delivery through GitLab/GitHub and JFrog Artifactory CI/CD pipelines.
- Collaborate on product operational processes for user training and certification, workspace, model and data-asset lifecycle management, and system disclaimers to ensure safety and compliance.
- Author and finalize architectural documents, LLM specifications, and installation-validation protocols.
- Create, update, and govern technical documentation in GitLab Markdown, Confluence, and GitLab Pages, including process-design documents.
- Facilitate collaboration across global and local cross-functional teams, communicating in Chinese and English.
- Provide architectural guidance and foster team capability growth.
- Experience in the pharmaceutical, life sciences, or top-tier technology industries, with exposure to large-scale compute platforms.
- Experience with architecture design relevant to GxP.
- Familiarity with at least one large-scale model training framework, such as Megatron-LM or MindSpore.
- Expertise in Ray for distributed training and inference pipelines.
- Experience with GCP Vertex AI, AWS Bedrock, Azure OpenAI, or Ali DashScope integration.
About the company
Roche is a global leader in pharmaceuticals and diagnostics. It develops medicines and diagnostic tools that help detect, monitor, and treat diseases, using research and biotechnology to drive new therapies and tests. Roche combines drug development with in-house diagnostic products to support personalized medicine, where treatments are tailored to individual patients based on test results. Its approach differs from many peers by integrating drug discovery with diagnostic capabilities and by expanding its research through strategic acquisitions (like Genentech) to strengthen its biotechnology and R&D capabilities. The company’s goal is to improve patient outcomes and healthcare worldwide by delivering reliable medicines, accurate diagnostic tests, and data-driven care.
Company Size
1-10
Company Stage
IPO
Headquarters
Basel, Switzerland
Founded
1896
Get referred to Roche
See people who can refer or advise you
Simplify's Take
What believers are saying
- Enicepatide delivered 15.5% weight loss in September 2026 and 22.7% in June.
- pTau217 can scale immediately through Roche's existing lab network and improve Alzheimer diagnostics.
- Roche's June 2026 data and September 2026 filings signal multiple obesity shots on goal.
What critics are saying
- Eli Lilly and Novo Nordisk dominate obesity pricing, distribution, and physician mindshare today.
- Enicepatide remains Phase II-only; Roche starts late-stage glycemic and cardiovascular studies in 2027.
- If enicepatide disappoints, Roche burns its $2.7 billion Carmot bet and loses obesity credibility.
What makes Roche unique
- Roche combines pharma, diagnostics, and installed cobas labs across 4,500 U.S. instruments.
- Elecsys pTau217 won FDA clearance on August 24, 2026, and CE marking September 7.
- Roche controls obesity assets from Carmot and Zealand, spanning GLP-1/GIP and amylin.
Help us improve and share your feedback! Did you find this helpful?
Benefits
Performance Bonus
Company News
The US Food and Drug Administration has approved Roche's Tecentriq in combination with a fluoropyrimidine and oxaliplatin for adjuvant treatment of stage III deficient DNA mismatch repair colon cancer. This marks Tecentriq's twelfth approved indication in the US. The approval follows the phase III ATOMIC study, published in The New England Journal of Medicine, which demonstrated that Tecentriq plus chemotherapy reduced the risk of disease recurrence or death by 50% compared to chemotherapy alone. The 36-month disease-free survival rate was 86% with the combination versus 76% with chemotherapy alone. Nearly 25% of colon cancers are diagnosed at stage III, and almost one in three stage III patients experience recurrence within five years. This represents the first approved adjuvant immunotherapy regimen for this patient population. Roche is pursuing further regulatory filings, including with the European Medicines Agency.
Roche posted strong Phase 2 results for its weight loss drug enicepatide, showing 22.5% mean weight loss over 48 weeks. This compares favourably to Eli Lilly's Zepbound, which achieved 20.2% weight loss over 72 weeks in a separate trial. In another Phase 2 study, enicepatide demonstrated HbA1c reductions of up to 2.65 percentage points in diabetic patients after 48 weeks, compared to tirzepatide's 2.37 percentage points after 52 weeks. Despite promising results, Roche faces stiff competition from Novo Nordisk and Eli Lilly, which both have deep pipelines in development. Novo Nordisk's CagriSema is nearing approval, whilst Eli Lilly's retatrutide has posted strong Phase 3 results. Roche's revenue grew 6% year over year in the first half of the year to 30.4 billion Swiss francs.
Roche has launched a newborn screening test that simultaneously detects three severe genetic conditions: Spinal Muscular Atrophy, Severe Combined Immunodeficiency Disease, and Sickle Cell Disease. TIB MOLBIOL, a Roche subsidiary, developed the LightMix Newborn TREC/SMN1/HBB kit for use in countries accepting the CE mark. The test enables early detection before symptoms appear, allowing clinicians to initiate life-changing treatments sooner. Early intervention can prevent severe nerve damage in SMA, enable lifesaving bone marrow transplants for SCID, and drastically lower infant mortality in SCD through preventive measures. The IVDR-approved test provides a ready-to-use solution that integrates into existing laboratory workflows. It runs on established LightCycler systems, making it accessible for private and academic hospital laboratories across Europe.
Roche's collaborator MediLink has announced positive phase III results for tambotatug pelitecan (Tam-Peli) in Chinese patients with relapsed small-cell lung cancer. The TAISHAN-302 trial showed Tam-Peli reduced the risk of death by 54% compared to topotecan, with median overall survival of 13.3 versus 9.4 months. The antibody-drug conjugate also demonstrated superior progression-free survival (7.4 versus 2.8 months) and objective response rates (59.1% versus 9.7%). China's National Medical Products Administration has accepted the New Drug Application for filing. Roche is developing and commercialising Tam-Peli outside mainland China, Hong Kong and Macau, with plans to rapidly initiate global trials. The therapy has received Breakthrough Therapy Designations from both the US FDA and China CDE for relapsed small-cell lung cancer.
Tempus AI has raised $600 million in Series C funding to scale its machine learning platform for cancer recurrence detection through liquid biopsies. The investment was led by SoftBank Vision Fund 2, with participation from Roche and Merck. Phase II trial data showed the AI-enhanced liquid biopsy analysis reduced false-negative results by 37% compared to conventional methods. The platform uses deep learning algorithms trained on over 12,000 genomic profiles to identify circulating tumour DNA fragments shed by residual cancer cells. A 2025 Nature Medicine study reported the system achieved 92% specificity in detecting minimal residual disease across breast, lung, and colorectal cancers. According to SEC filings, 45% of funds will support FDA pre-market approval trials, whilst 30% will fund global expansion into EU and APAC markets.