Averroes.ai
Website:
averroes.ai
Company:
https://www.linkedin.com/company/averroes-ai-inc
Industries: Semiconductor Manufacturing
Job details:
About Averroes.ai
Averroes.ai is an AI-powered visual inspection platform serving semiconductor fabs, photomask manufacturers, electronics/PCB, automotive, and advanced manufacturing companies. Our flagship product, WatchDog, uses a cascading supervised and unsupervised architecture to deliver near-zero defect escapes without requiring hardware changes. We are a pure-software, hardware-agnostic solution that layers on top of existing inspection equipment (KLA, Camtek, Onto Innovation, and others).
The Opportunity
We are building a new capability at the intersection of AI and Advanced Process Control (APC). Today, Averroes.ai tells customers what went wrong in their inspection process. This role will architect and own the intelligence layer that tells them what is about to go wrong and when their equipment needs tuning.
This is a senior individual contributor position with significant autonomy. You will define the technical architecture, select the modeling approaches, and own the end-to-end delivery of production-grade predictive systems. You will work directly with machine telemetry logs, equipment tuning records, and measurement data from real customer environments to build models that detect drift before it impacts yield.
This is not a research role and not a dashboard role. You will ship models to production, iterate based on real-world feedback from process engineers, and own the outcomes.
What You Will Own
Predictive Equipment Tuning (Primary)
- Architect and build the ML pipeline that predicts when manufacturing equipment requires recalibration or tuning, using time-series analysis of machine logs, sensor telemetry, and historical tuning records. You will make the design decisions on model architecture, feature engineering, and inference strategy.
- Design and own measurement drift detection systems that identify gradual tool degradation before it crosses specification limits. This requires combining statistical process control (SPC) methods with modern ML techniques and making the right trade-offs between sensitivity and false alarm rates.
- Build anomaly detection for real-time equipment health monitoring, classifying tool states (normal, degrading, out-of-spec) from multivariate sensor streams. You will determine the right approach for each customer environment, whether that is isolation forests, autoencoders, change-point detection, or something else entirely.
- Create remaining-useful-life (RUL) and time-to-next-tuning models that enable predictive maintenance scheduling. The goal is reducing both unplanned downtime and unnecessary preventive maintenance, and you will own the metrics that prove it.
- Design and implement concept drift and data drift detection to ensure deployed models stay accurate as process conditions, recipes, and equipment configurations change over time.
APC Intelligence Layer
- Define the architecture for integrating predictive tuning signals with Averroes.ai's existing inspection data. The end state is a closed-loop feedback system that links defect patterns to upstream equipment behavior.
- Design fault detection and classification (FDC) models that correlate equipment parameter deviations with downstream quality outcomes. You will work directly with customer process engineers to validate these correlations.
- Build interpretable model outputs that equipment engineers can trust and act on: confidence intervals, feature importance, root-cause indicators, and actionable tuning recommendations. The end users are process engineers, not data scientists.
Technical Leadership
- Set the technical direction for Averroes.ai's APC and predictive analytics capability. You will evaluate and select the tooling, frameworks, and infrastructure patterns that the team builds on.
- Define data contracts and integration specifications for customer equipment data ingestion. You will determine what data we need from customers, in what format, and how it flows through our systems.
- Contribute to customer-facing technical discussions, proof-of-concept scoping, and solution architecture during the pre-sales and onboarding process. Your domain expertise is a competitive advantage.
- As the team grows, mentor junior data scientists and ML engineers. Establish coding standards, model validation practices, and deployment protocols for the APC product line.
Computer Vision (Secondary)
- Contribute to WatchDog's defect detection models where process drift manifests as subtle visual changes in inspection images, bridging the gap between APC signals and CV outputs.
- Collaborate with the existing CV team on model improvements and edge-case handling. This is a supporting responsibility, not a primary one.
Required Qualification
- MS or PhD in Data Science, Machine Learning, Electrical Engineering, Industrial Engineering, Statistics, or a related quantitative field.
- 5+ years of hands-on experience building and deploying ML models on time-series and sensor data in manufacturing, semiconductor, or industrial environments. At least 3 of those years should involve direct work with equipment or process data in a production setting.
- Demonstrated ownership of at least two of the following in a production environment: fault detection and classification (FDC), statistical process control (SPC), advanced process control (APC), predictive maintenance systems, or virtual metrology.
- Strong proficiency in Python with production experience using PyTorch or TensorFlow, plus time-series libraries (statsmodels, tslearn, Prophet, ARIMA-family models, or equivalent).
- Deep experience with anomaly detection methods applied to multivariate sensor data: isolation forests, autoencoders, change-point detection, CUSUM, or similar. You should be able to explain the trade-offs between these approaches and know when each is appropriate.
- Hands-on experience with drift detection techniques (both concept drift and data distribution drift) and strategies for maintaining model accuracy in non-stationary environments.
- Proven ability to work with messy, high-volume, real-world equipment logs: handling missing data, variable sampling rates, mixed schemas, and noisy sensor signals. We will ask for specific examples.
- Track record of building models that non-ML stakeholders trust and use. You should be comfortable presenting to process engineers, explaining model behavior, and translating statistical concepts into operational decisions.
- Experience architecting ML systems end-to-end: from data ingestion and feature engineering through training, validation, deployment, and monitoring. You do not hand off notebooks to an engineering team.
What Sets You Apart
We are looking for someone who has already solved these problems at least once in a real environment. You have wrestled with equipment data in production: logs that change format across tool vendors, tuning records stored in inconsistent formats, sensor signals that drift differently across chambers on the same tool, and measurement data where the noise floor is close to the signal.
You know the difference between a model that performs well in backtesting and one that process engineers actually use to make tuning decisions. You have built that trust before, and you can articulate how.
You are comfortable operating with ambiguity. We have customer demand and real data, but we do not have a playbook for this product. You will write the playbook.
Why Averroes.ai
- Greenfield product area with validated customer demand: this is not speculative R&D. Customers are asking for this capability and providing data for it.
- Your models will run in production at semiconductor fabs and advanced manufacturers. This is not a proof-of-concept role.
- Small, senior team with direct access to customers, real-world data, and the CEO from day one. Zero organizational layers between your work and business impact.
- Significant ownership and autonomy. You will define the technical roadmap for this product line, not execute someone else's.
- Competitive salary, equity, and benefits. Remote-friendly with preference for APAC time zones.
Interview Process
- Introductory call with CEO (30 min): mutual fit, role expectations, your background in APC/predictive maintenance.
- Technical deep-dive (60 min): walk us through a past project where you built and deployed a predictive model on equipment or sensor data. We will ask about architecture decisions, trade-offs, and what you would do differently.
- Take-home exercise (4-6 hours): we provide a sample equipment dataset with tuning logs. Build a drift detection or tuning prediction model. Present your approach, results, and productionization plan.
- Final conversation (45 min): team fit, working style, questions from both sides.
Click on Apply to know more.