Machine Learning Engineer

As a Machine Learning Engineer at Reducto, you will be tasked with training and deploying state-of-the-art machine learning models for parsing complex documents. You'll experiment with new techniques to enhance LLM accuracy, build data pipelines, evaluate model performance, and integrate models into our product. Collaborating with our founders and customers will also play a crucial role in shaping the direction of our engineering strategy.

To apply for the Machine Learning Engineer role at Reducto, you should have 2 to 5 years of experience in training, fine-tuning, and evaluating ML models used in production. Proficiency in Python and familiarity with traditional computer vision and vision-language models (VLMs) are essential skills. Additionally, a proactive approach to problem-solving and the ability to iterate quickly would be beneficial.

In the Machine Learning Engineer position at Reducto, you'll utilize a variety of tools aligned with machine learning, such as Python for model development. You may also build additional tools, such as Streamlit applications for hypothesis testing and data visualization. Familiarity with LLMs, traditional ML techniques, and data pipeline tools will also be crucial for success in this role.

Yes, the Machine Learning Engineer role at Reducto is an in-person position based in our San Francisco office. This setup allows for better collaboration with the founding team and quicker iterations on projects, which is vital for an early-stage company where working hard and moving fast is essential.

The company culture at Reducto is one that values high standards and rapid execution. As a Machine Learning Engineer, you'll be part of an innovative team where appreciation for quality work and proactive problem-solving is paramount. You'll have the freedom to take initiative, and your contributions will directly shape product development as we all work together toward a common goal.

To train machine learning models in production, I start with understanding the business problem and gather a well-defined dataset. I prioritize data preprocessing and cleaning to ensure quality inputs. Subsequently, I select appropriate algorithms and split data into training and testing sets. Continuous evaluation of model performance using metrics and testing on real-world scenarios is essential for effectiveness before deployment.

My experience with fine-tuning LLMs involves customizing pre-trained models to specific tasks. I analyze the domain-specific data, adjust hyperparameters, and use techniques like transfer learning to improve the model's performance. I focus on monitoring model performance throughout the fine-tuning process, making adjustments as necessary based on feedback and validation results.

In a previous project, I successfully deployed a machine learning model by thoroughly testing its performance in a staging environment. I closely monitored resource usage and response times after deployment, making optimizations based on real-world data. Collaboration with cross-functional teams ensured that the deployment was smooth, and I used tools for continuous monitoring to quickly identify and address any issues.

I frequently use tools like Apache Airflow and Luigi for orchestrating data pipelines. In addition, I leverage Python libraries for data manipulation, such as Pandas and NumPy, and integrate with platforms like AWS or Azure for storage and computational capabilities. It’s all about designing scalable and reliable pipelines to ensure seamless data flow.

I evaluate model performance by using metrics like accuracy, precision, recall, and F1 score, depending on the context of the problem. I also consider creating confusion matrices to visualize performance across various classes. Additionally, I implement A/B testing when feasible to compare model variants in production, helping ensure we choose the best performing model.

Handling unstructured data requires a different strategy compared to structured data. My experience includes leveraging NLP techniques for text extraction and using computer vision for processing images. I focus on building custom preprocessing steps that help convert unstructured data into a more usable format. I also believe in using heuristics to design pipelines that effectively manage this complexity.

I stay updated with AI and ML developments by following influential research papers, attending webinars, and participating in community discussions. I also subscribe to newsletters and online courses to continuously learn new techniques and tools. Networking with industry peers helps me exchange insights and best practices, which in turn fuels innovation in my work.

In a recent project, I encountered an issue where the model's predictions were inconsistent. I systematically debugged by analyzing the input data, checking for anomalies, and ensuring that preprocessing steps matched the training phase. I also revised the training process to identify if any hyperparameters needed adjustment, leading to improved model accuracy post-debugging.

Collaboration is critical in machine learning projects, as complex problems often require insights from diverse perspectives. Interaction with data scientists, software engineers, and product managers ensures alignment of goals and innovative solutions. Furthermore, collective problem-solving can lead to faster iteration and improved product outcomes, as everyone brings unique skills to the table.

I am passionate about using AI to solve real-world problems, and Reducto’s mission to process unstructured data aligns perfectly with my interests. I admire the innovation in your approach and I’m excited about the opportunity to collaborate closely with a founding team and contribute to building impactful products that save time and enhance productivity for enterprises.