What you will do

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.

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.

1份Architect工作在Seremban - October 2026

AI Architect