Founding AI Engineer

About the Role

TaloTrace is developing AI agents designed to test software on real devices. These agents navigate web applications and mobile builds on platforms like web, Android, and iOS, whether using emulators or physical devices, to identify bugs and report issues effectively. As a compact, driven team that values rapid iteration, TaloTrace seeks a Founding AI Engineer to take ownership of the model layer powering these agents. This role involves managing models that analyze screens, determine actions, execute them, and assess outcomes, striving to make each model smaller, faster, more cost-efficient, and more accurate.

Key Responsibilities

• Manage the entire lifecycle of the model layer, including data management, training, assessment, deployment, and operational cost monitoring.
• Translate research developments into production-ready solutions and discontinue ineffective models.
• Develop evaluation systems that accurately reflect performance, including identifying failures.
• Balance and quantify trade-offs between cost, quality, and latency.
• Establish standards for modeling work: defining measurement criteria, validation for improvements, and requirements before deployment.

Potential Areas of Ownership

• Developing smaller models through methods like distillation, fine-tuning, and quantization to reduce computational demands.
• Improving perception and grounding for consistent element localization across platforms, even after interface redesigns.
• Advancing judgment models to evaluate the validity and severity of findings and determine uncertain outcomes.
• Designing exploration policies to simulate realistic user navigation beyond typical usage paths.
• Implementing state prediction to anticipate screen changes post-actions, aiding navigation and anomaly detection.
• Creating evaluation infrastructure including offline benchmarks and regression checks to validate models.
• Optimizing inference efficiency via quantization, batching, caching, and facilitating on-device deployment while managing cost-quality balance per service tier.
• Constructing data pipelines that handle collection, cleaning, labeling, and versioning of interaction datasets.

Qualifications and Experience

• At least four years' experience in machine learning or AI engineering, or substantial software engineering experience with hands-on model ownership and deployment.
• Proficient in Python and experienced with PyTorch or comparable frameworks; capable of implementing published research repositories.
• Experience in training and refining models using techniques like supervised fine-tuning, LoRA/PEFT, reinforcement learning, behavior cloning, or distillation.
• Skilled in building robust evaluation setups and discerning reliable metrics.
• Practical exposure to large language model systems, including structured outputs, tool integration, understanding failure cases, and managing latency and costs.
• Competent with data pipeline creation involving data collection, preprocessing, and version control.
• Adherence to rigorous validation of results, identifying when outcomes are unsuccessful.
• Able to rapidly prototype experiments for investigation and discard them after use.
• Strong communication skills, able to document experiments comprehensively for team use.
• Effective use of AI coding aids, producing work that withstands code review.

Desirable Additional Skills

• Startup experience as an early engineer familiar with delivering initial product versions.
• Familiarity with vision-language models and multimodal training approaches.
• Background in graphical user interface agents, computer interaction models, or robotics and embodied policies.
• Experience with imitation learning or behavior cloning using human demonstration datasets.
• Expertise optimizing model inference with techniques like FP8 or INT4 quantization, vLLM, TensorRT, caching strategies, or on-device execution.
• Applied reinforcement learning methods such as RLHF, GRPO, or reward verification mechanisms.
• Modeling human behavior across domains including game AI, recommendation systems, user simulations, or behavioral biometrics.
• Contributions to research through publications, open source projects, or reproducibility efforts in related fields.

Recruitment Process

• Stage 1: Seven-day coding challenge.
• Stage 2: Interview with the CTO.

Company Culture and Values

At TaloTrace, the commitment is to "leave quality to us." The mission is to empower builders and innovators to bring software from conception to delivery without lingering doubts about functionality. The environment fosters quick, empirical decision-making through experimentation and open debate, with expectations for everyone to contribute across different areas beyond their core expertise.

Additional Information

The position is based in Singapore and operates onsite with hybrid flexibility. Candidates are welcomed from diverse backgrounds, and accommodations during the hiring process are available upon request.