Compensation: $60k – $72k SGD

Senior Applied AI Engineer (Agents & Evaluation)

Gengis AI Location: Singapore (onsite preferred; hybrid or remote considered) Employment: Full-time Salary: S$60,000–S$72,000 / year

About Gengis

Gengis builds AI for film and TV production. Our platform, YAM, connects work from script to screen, carrying production context, creative decisions and approvals through the process.

Our founders also built Refinery Media and X3D Studio. YAM grew out of work on real productions, and you’ll work directly with the filmmakers and engineers building it.

The role You’ll own two areas of our AI development, with your time split evenly between them:

Model development and fine-tuning (50%). Continue developing, training, evaluating and improving the models built through our collaboration with AI Singapore. AI agents and evaluation (50%). Design, build and optimise the agents that power YAM, so they deliver reliable, measurable results at a practical cost.

You’ll work across machine learning and applied AI systems. You’ll take these models beyond the initial research and development stage, and your remit will grow into production agent systems, model selection, evaluation infrastructure and AI product engineering.

You’ll work closely with our engineering team and with film practitioners, turning creative requirements into reliable AI capabilities.

What you’ll own: model development and fine-tuning (50%) You’ll take ownership of the performance, reliability and ongoing development of the models built during the AISG project.

Continued fine-tuning. Improve the existing models through supervised fine-tuning (SFT), experimentation and changes to training configuration, building on the work already done. Dataset development. Build and refine training and validation datasets, improve data quality, find weaknesses and manage dataset versions. Model evaluation. Set benchmarks for accuracy, consistency and generalisation, and compare performance before and after fine-tuning. *Experimentation. * Test training approaches, hyperparameters and data strategies to improve performance while keeping compute costs under control. Inference optimisation. Improve serving efficiency, latency and resource use without lowering output quality. Production integration. Work with engineering to bring improved models into YAM and keep them running reliably in production workflows. Continuous improvement. Investigate model failures, gather feedback, find performance gaps and feed what you learn into the next round of training.

We are not training a foundation model from scratch. The goal is to keep improving the models we have and make them useful, reliable and maintainable in production.

What you’ll own: AI agents and evaluation (50%) You’ll design and improve the agents behind YAM, with a focus on measurable quality, reliability and cost.

*Agent design and development. * Build and configure agents, including their prompts, skills, tools, context retrieval, structured outputs and execution workflows.

*Evaluation frameworks. * Create evaluation datasets, scoring rubrics, golden fixtures and regression tests that measure agent quality consistently. Model selection. Compare LLMs across providers such as OpenRouter and AWS Bedrock on output quality, latency and cost. Prompt and context optimisation. Experiment with prompt strategies, reasoning effort, tool configurations and context selection to improve agent performance. *Human-aligned evaluation. * Work with film practitioners to understand what good output looks like, and turn their feedback into measurable criteria. Production monitoring. Investigate failed runs, hallucinations, inconsistent outputs and rising costs, and turn what you find into concrete improvements. Engineering delivery. Implement improvements directly in our codebase, write tests and contribute to production releases.

Questions you’ll help answer Did fine-tuning actually improve the model’s performance? Should an agent use our fine-tuned model or a third-party LLM? Can we reduce inference costs without reducing output quality? Does a different prompt or agent configuration give more reliable results? How do we measure creative AI output against what practitioners expect? We expect these questions to be answered with reproducible experiments, not assumptions.

What you’ll work on first Your first priorities build directly on the work already completed with AI Singapore.

Model development. Take over development of the existing fine-tuned model. Review how it performs today, identify where it can improve and set up a repeatable workflow for training, evaluation and deployment.

Script scoring and story breakdown. Improve the quality and consistency of YAM’s script scoring and story breakdown. Set benchmarks using labelled examples and filmmaker-approved criteria, then compare models, fine-tuning approaches and agent configurations.

*Writer’s Room agents. * Apply these foundations to our next generation of creative AI tools: scene drafting from story outlines, writing proposals generated by multiple models, character interviews that respect narrative context, audience reaction simulations and shot lists generated from reviewed scripts. For each capability, you’ll establish a quality baseline and measure the cost and latency of producing an acceptable result.

What you’ll bring You have practical experience in machine learning and a strong interest in building production AI systems.

Hands-on experience with machine learning and deep learning, particularly LLMs and fine-tuning workflows.

Experience preparing training datasets, running model experiments and evaluating model performance.

Strong Python, including PyTorch, Hugging Face Transformers or similar ML frameworks.

An understanding of LLM inference, prompting, structured outputs and basic agent workflows.

Familiarity with evaluation methods, performance metrics and systematic experimentation.

The ability to write maintainable code, work with Git and collaborate within an existing codebase.

Strong problem-solving skills, and a willingness to investigate failures at both the model and the system level.

Strong spoken and written English. You’ll work regularly with our engineering team and with film industry practitioners.

Useful additional experience Supervised fine-tuning (SFT), LoRA or QLoRA. Multimodal models, information extraction or document understanding. Agent frameworks, tool calling, MCP or retrieval-augmented generation (RAG). LLM-as-judge, human evaluation, rubric design and benchmark development. Model serving, GPU optimisation, Docker, cloud infrastructure or MLOps. TypeScript, PostgreSQL, APIs or backend development. Screenwriting, creative AI, or film and television workflows.

You don’t need to know our whole stack. We value practical engineering ability, curiosity and a willingness to learn.

How you’ll work You’ll work across model experimentation and production software development. We expect short, focused experiments with clear objectives, documented findings and regular checkpoints.

For model work, you’ll keep the datasets, model versions, training configurations and evaluation results needed to reproduce each experiment. For agent work, you’ll track prompts, model configurations, inputs, outputs, latency, cost and quality metrics.

Filmmakers define creative quality. You take engineering ownership of how we measure and improve AI performance.

A well-documented negative result is still valuable. We want someone who can explain what worked, what didn’t and what we should try next.

What success looks like after six months The AISG model has a clear owner, a reproducible training pipeline and a measurable improvement roadmap. Fine-tuning experiments are tracked and evaluated against consistent benchmarks. The model runs in production workflows, with monitoring and regression testing. Key YAM agents have documented configurations, evaluation criteria and measurable quality baselines. Model and agent choices are backed by evidence on quality, latency and cost. The team can say with confidence whether a model update, prompt change or agent change actually improved the product.

Apply

Send your CV or profile and a short example of an agent system you shipped or an evaluation that changed a product decision. Tell us what you owned, how you measured the result and what you learned. A written summary is welcome if the work is confidential.

Apply through Wellfound.