Product Manager (Training)

Fireworks AI; United States; On-site; USD 120,000 - 200,000; Full time

Job summary

Fireworks AI is hiring a Product Manager for its training product, focusing on strategy, specs, and customer-facing tuning work. You will collaborate across the training, research, and inference teams to drive real customer impact and scale self-serve and enterprise training capabilities.

You will own roadmaps and success metrics for API, UI, and CLI components, while partnering with product marketing, sales, and field to define pricing, packaging, and enablement efforts.

Qualifications

• Experience with GPU cost, throughput and quality tradeoffs.
• Excellent written communication for specs, launches, and customer explanations.
• Open-source or developer-community background preferred.
• Strong technical foundation in CS or Electrical Engineering.
• Experience with ML platforms, MLOps, or AI infrastructure products.

Responsibilities

• Own the roadmap, strategy, and success metrics for parts of the training product across API, UI, and CLI.
• Write specs, sit with customers running real tuning jobs, and work with training/research/inference teams.
• Help grow self-serve usage and simplify training for enterprise users.
• Partner with marketing, sales, and field to land pricing, packaging, docs, and enablement.

Skills

GPU economics

Written communication

Open-source experience

Technical background (CS/EE)

ML platform / MLOps experience

Reinforcement fine-tuning familiarity

Startup/founding experience

Product ownership (2-8+ years)

End-to-end product ownership

Ambiguity tolerance

Hands-on model fine-tuning

Education

CS/EE degree

Tools

ML Platform

MLOps

Model Registries

Eval Tooling

Job description

• Fireworks is building specialized intelligence to enable companies to own their AI with frontier-beating quality and efficiency
• Training is at the core of how they get there
• As a PM working on training, you’ll help set strategy, write specs, sit with customers running real tuning jobs, and work across our training, research, and inference teams to deliver real customer impact
• Example problems you may work on include determining how to grow our self-serve training usage or making training easier and faster for our enterprise users
• Own the roadmap, strategy, and success metrics for parts of the Fireworks training product, across API, UI, and CLI
• Work directly with AI-native startups and enterprises running real training workloads — watch them work, unblock them whey they stall, and convert repeated pain into productized capability
• Turn the bespoke work our forward-deployed and applied ML teams do for top accounts into scalable, self-serve product
• Partner with product marketing, sales, and the field to launch training capabilities that land — pricing and packaging, docs, cookbooks, and enablement included

PREFERRED QUALIFICATIONS:

Understanding of GPU economics and how training cost, throughput, and quality trade off against each otherExcellent written communication. You can write a spec, a launch post, and a customer-facing explanation of a tradeoff, and all three will be clearOpen-source or developer-community experience — you know what makes an SDK or API feel good to useStrong technical background — CS/EE degree, production engineering experience, or equivalent depth earned on the jobExperience with ML platform, MLOps, or AI infrastructure products — training platforms, eval tooling, or model registriesFamiliarity with reinforcement fine-tuning specifics: reward modeling, rollout environments, and agent-training workflowsEarly startup or founding experience2 – 8+ years of product management experience building technical or developer-facing products (we are hiring at multiple levels for this role)Familiarity with the post-training lifecycle: dataset curation, SFT, LoRA/PEFT, RL-based methods, evaluation, and how these connect to inference in productionComfort with ambiguity, and a bias toward shipping and learning over waiting for certaintyDemonstrated ownership of a product area end to end from strategy, spec, launch, to metricsDeep hunger and motivation. This isn’t a 9-5 job and you’ll be expected to step up, especially during periods of “wartime.”You’ve personally fine-tuned models and shipped the result into something real