Staff Machine Learning Engineer, Reddit

• Deep ML Expertise: Hands-on experience with tree models, neural nets, ranking, and especially LLM/VLM fine-tuning & RLHF/RLPF for ads
• Quant & Experimentation Chops: Comfort with uplift metrics, variance, minimum detectable effects and experimentation setups for generative systems
• Ads / Marketplace Insight: Understanding auction dynamics and advertiser pain points to create tools that drive revenue and retention
• Startup Mindset: Bias to action, comfort with ambiguity, and scrappiness: ACE evolved from a startup acquisition and still moves fast
• Communication Superpowers: Able to distill complex ML trade-offs for executives, product managers, and non-technical stakeholders alike
• Prior work on ad creative tooling: Headline generators, image generators, pre-test models, editing flows for the advertiser

Desirable

• Reddit’s newly minted Ads Creative Effectiveness team sits at the heart of our creative strategy, building GenAI and predictive products that help advertisers ship high-impact campaigns
• We are searching for a Staff engineer with in-depth GenAI expertise that can help us build the future of generative products for ad creative at Reddit - best-in-class image, video and copy generators that leverage Reddit’s best practices, performance-guided generations that maximize clicks and conversions, and brand safe, product preserving image/video editing pipelines
• Advanced Generative Workflows: Architect and implement state-of-the-art pipelines for ad creative generation. You will move beyond vanilla prompting to build systems capable of product, font, logo and brand asset preservation using techniques like IP-Adapters, and LoRAs
• End-to-End Image & Video Editing: Build editing capabilities allowing advertisers to smart-crop, inpaint, outpaint, resize, and style-transfer assets while maintaining strict brand guidelines
• LLM/VLM Fine-Tuning & RAG: Adapt large vision-language models to understand "winning" ad aesthetics. You will build feedback loops where ad performance metrics (CTR, CVR) are used to fine-tune the generation process (RLHF/RLPF for Ads)
• Production-Grade Serving: Design low-latency inference stacks for diffusion and transformer models. You will solve for the trade-off between generation quality, cost-per-impression, and serving speed
• Safety & Compliance Layer: engineer robust filtering and safety systems. You ensure our models never generate non-compliant, unsafe, or copyright-infringing content
• Partner Broadly: Collaborate with Product, Sales, Policy, UX, and sibling Ads ML teams to integrate ACE capabilities into the Ad Manager UI and the ad delivery path
• Guardrails & Safety: Embed brand-safety filters, copyright checks, and human-in-the-loop escalation paths to protect users and advertisers