Founding ML Engineer, Post-Training

Aghanim is an integrated commerce, liveops automation, community engagement, and payments platform for video games.

Mobile games have traditionally depended on app stores for distribution, payments, and player relationships. We believe there is a better way. Aghanim helps game studios build direct relationships with players, sell directly, and build their future on their own terms. Today, more than 100 games worldwide are already building this future with Aghanim.

Our team brings together people across Los Angeles, New York, Seoul, Beijing, London, Lisbon, Belgrade and other locations around the globe, with deep expertise in gaming, fintech and technology. We move quickly, keep communication direct, and focus on getting things done. We believe the best people thrive when they have autonomy, ownership, and a stake in the company's success.

We’re looking for a Founding ML Engineer at Senior or Principal level to lead the transition of Newton, Aghanim’s AI product for partners, from proprietary model APIs to models we control.

Newton already runs in production on frontier models. You’ll own the model layer end to end: data collection, evaluation, post-training, serving, and rollout. The goal is to preserve the quality partners rely on while reducing cost, improving speed, and keeping partner data within our infrastructure.

You’ll work closely with the product to understand its behavior and partner expectations, then independently define the technical approach and priorities. Decisions about training, routing, and model replacement should be backed by production data and evaluations.

• Build production data pipelines that capture versioned instructions, tool calls, outputs, costs, and outcomes, with anonymization, partner-level deletion, and expert corrections.

• Develop benchmarks, regression tests, and code-based and LLM-judge evaluators using real Newton runs to measure factual accuracy, tool use, answer quality, cost, and latency.

• Compare frontier and open-weight models, route tasks to suitable models, and identify where prompting, distillation, or post-training delivers the best results.

• Train and adapt models using distillation, SFT, LoRA/QLoRA, and DPO on accepted trajectories and expert feedback; use RL where simpler approaches fall short.

• Deploy models behind Newton’s existing interface, validate them through shadow runs, and manage rollout based on quality benchmarks and serving economics.

Tech stack: Python, TypeScript, SQL, self-hosted Langfuse, TRL / Unsloth / Axolotl or managed training platforms, vLLM, serverless GPUs, GCP / Modal; open-weight models including Qwen, Gemma, DeepSeek, and GLM.

• Hands-on experience post-training LLMs and shipping them to production using SFT, LoRA/QLoRA, DPO, and distillation.

• Experience building evaluation systems from production data, including offline benchmarks, regression and behavior tests, and code-based and LLM-judge evaluators.

• Understanding of agentic systems, including tool calling, multi-step trajectories, structured outputs, and the impact of changing the underlying model.

• Practical experience serving models with vLLM or similar tools, quantization, and GPU deployment, including latency, cold-start, and cost tradeoffs.

• Strong production Python and SQL skills, with enough TypeScript proficiency to modify product code and instrumentation independently.

• Ability to understand product behavior and user expectations, translate them into measurable quality requirements, and decide when training is justified.

• Experience independently owning a technical area, setting priorities, and working across an unfamiliar codebase and its design documents.

• Regular use of AI coding agents, with a clear approach to verifying their output.

• Experience migrating a production product from proprietary model APIs to self-hosted or open-weight models.

• Experience with RL for LLMs, particularly GRPO with automatic verifiers.

• Familiarity with Langfuse, Braintrust, or similar tracing and evaluation tools.

• Experience with GCP, including Vertex AI, Cloud Run GPU, and BigQuery.

• Interest in game monetization and analytics.

• World-class team – work alongside experienced professionals from around the globe who have built products used by millions of players

• High growth, high impact – be part of a fast-growing company where ideas turn into products and reach customers in days, not months

• Autonomy and ownership – we trust people to make decisions, take initiative, and drive results

• Modern tools and technology – use AI, automation, and modern tools as part of your everyday work

• Equity – participate in the company's growth and long-term success