AI Architect
Two95 International Inc.
·Today
- Location
- Kuala Lumpur, Federal Territory of Kuala Lumpur, Malaysia
- Workplace
- Onsite
- Type
- Full-time
- Department
- Engineering
- Experience
- 8+ years
- Source
- Workable
Description
ROLE: AI Architect
JOB TYPE: FTE
LOCATION: Seremban,
JOB DESCRIPTION:
We are looking for AI Architect with 8+ years of experience.
Job responsibilities
Design the flow and architecture
- Map the current lot disposition flow with quality, product and process engineers (hold triggers, evidence gathered, decision rules, MRB, approval levels) and design the target AI-assisted flow: what is automated, what is assisted, and what stays manual.
- Own the end-to-end solution architecture: data ingestion, feature and context layer, ML models, LLM/agent reasoning, decision workflow and audit trail.
- Produce the High-Level Design (HLD): system context, component, data-flow and deployment views, integration landscape, and non-functional requirements such as latency, availability, security and scalability.
- Produce and sign off the Low-Level Design (LLD): service and API contracts, data models and schemas, agent and tool definitions, model and prompt interfaces, sequence diagrams and error handling.
- Make and document the key trade-offs (build vs buy, cloud vs on-prem, model selection) and own the architecture decision records.
Build the models and agents
- Build agents that assemble the evidence pack for a held lot by pulling lot genealogy, SPC charts, inline and parametric data, wafer maps, test-bin summaries and equipment history through tool-calling.
- Combine traditional ML with LLM reasoning: hold-reason classification, anomaly and excursion detection, wafer-map pattern recognition, and similarity search over historical dispositions and 8D/MRB records.
- Generate disposition recommendations with confidence scores, cited evidence and a plain-language rationale that an engineer can verify in seconds.
Set best practices for tooling and engineering
- Define the team's engineering toolchain and standards: repository structure, coding standards, branching and code review, CI/CD, infrastructure as code and environment strategy.
- Set MLOps/LLMOps practices: experiment tracking, model registry, prompt versioning, evaluation pipelines, monitoring, drift detection, rollback and cost tracking.
- Set documentation standards (HLD/LLD templates, ADRs, runbooks) so the system can be supported and extended beyond the original team.
Make it safe for automotive-grade quality
- Design the human-in-the-loop controls: confidence thresholds, approval routing, override capture, and a staged path from recommend-only to auto-release for well-understood, low-risk hold types.
- Build full traceability into every recommendation (input data, model and prompt version, rationale and final human decision) to meet IATF 16949 and customer audit expectations.
- Agree the evaluation bar with quality owners before anything goes live: back-testing against historical dispositions, false-release and escape-risk metrics, drift monitoring and rollback.
Lead and guide the team
- Lead the AI engineers on the programme: break the HLD and LLD into work packages, plan delivery with the team, and unblock design questions as they come up.
- Walk engineers through the design before development starts, run design and code reviews, and make sure what is built matches the architecture.
- Mentor the AI engineers toward independent ownership of their components.
- Pilot at one site, prove the impact on hold cycle time and engineering effort, then turn it into a reusable pattern for rollout across fabs and assembly & test sites.
Job requirements
Core requirements
- 8+ years building ML/AI systems, with at least 2 years hands-on with LLM or agent systems.
- Has architected and shipped an AI decision-support or automation system to production, not just a prototype.
- Proven track record of writing HLD and LLD for production systems and leading system integration across enterprise or manufacturing systems.
- Experience leading or technically guiding a team of engineers from design through development and release.
- Strong traditional machine learning: classification, anomaly detection, time-series and tabular models (e.g., gradient boosting), feature engineering, and evaluation on imbalanced data.
- Hands-on with agent frameworks or SDKs (Claude Agent SDK/MCP, LangGraph, AutoGen) or custom orchestration, and with RAG over both structured and unstructured data.
- Strong Python and SQL; comfortable with large volumes of tabular, time-series or sensor data.
- Comfortable defining engineering best practices and tooling: CI/CD, code review, testing strategy and MLOps.
- Experience designing human-in-the-loop systems where explainability, auditability and safe failure matter.
- Able to sit with process and quality engineers, understand how they actually decide, and turn that into an architecture they trust.
Good to have
- Semiconductor manufacturing experience, front-end fab or back-end assembly & test, and familiarity with lot disposition, SPC, yield management, wafer maps or test data (e.g., STDF).
- Experience integrating with MES and other manufacturing systems: SPC, yield management, equipment and FDC data.
- Computer vision for inspection images (AOI, SEM, visual defect classification).
- Architecture modelling and documentation (e.g., C4, UML, sequence diagrams) and integration patterns such as REST, event streaming and message queues.
- Cloud AI platforms (Azure AI Foundry, AWS Bedrock/SageMaker), containers (Docker, Kubernetes) and CI/CD tooling.
- MLOps/LLMOps tooling such as MLflow, LangSmith or Arize: model registries, eval pipelines, drift monitoring and cost tracking.