Senior ML Engineer
Social Discovery Group3 hours ago
Remote, SerbiaSenior
Responsibilities
- Set the ML direction by selecting strategic bets and connecting them to business needs.
- Drive LLM post-training and model adaptation from dataset construction through fine-tuning and preference optimization.
- Design and build agent harnesses, tool-using LLM systems, and multi-agent development workflows.
- Build evaluation frameworks and LLM-as-a-judge systems that predict real-world performance.
- Improve dialogue quality across memory, emotional intelligence, persona consistency, safety, and anti-loop behavior.
- Ship production systems that influence user experience, retention, and monetization.
- Implement production LLM integrations, including APIs, streaming, batching, fallbacks, and cost/latency trade-offs.
- Build observability for agents through traces, tool-call logs, transcript analysis, and failure clustering.
Requirements
- Expert-level Python proficiency, including clean code, typing, asynchronous programming, multiprocessing, profiling, debugging, notebooks, and production code.
- Experience training and fine-tuning LLMs, including SFT, LoRA/QLoRA, PEFT, and preference optimization methods such as DPO, ORPO, SimPO, CPO, or GRPO.
- Experience building LLM training datasets through cleaning, filtering, deduplication, preference-pair creation, synthetic data, and hard-negative mining.
- Experience with offline and pairwise evaluation, LLM judges, and human annotation.
- Experience designing agent harnesses with tool calling, function calling, structured outputs, retries, routing, state management, memory/context, and guardrails.
- Strong understanding of transformer architecture and modern LLM behavior.
- Accuracy with data types, devices, batching, fp32, fp16, bf16, quantization, CUDA memory, and dtype mismatch debugging.
- Experience with Git, Docker, Linux, and basic MLOps.
- Fluent Russian and readiness to work within CET ±2 hours.
- Preferred: experience with computer vision, image generation, multimodal systems, long-running conversations, character consistency, vLLM, TGI, SGLang, DeepSpeed, FSDP, Accelerate, quantization, and safety classifiers.
Benefits
- Fully remote, full-time opportunity.
- 28 calendar days of vacation per year and 7 wellness days per year.
- Bonuses of up to $5000 for recommending successful applicants.
- 50% payment for professional training, international conferences, and meetings.
- Corporate discount for English lessons.
- Health benefits, including up to $1000 gross per year reimbursement for eligible employees without corporate medical insurance.
- Equipped workplace and necessary equipment in offices or coworking locations, or reimbursement of workplace costs up to $1000 gross once every 3 years.
- Internal gamified gratitude system with merchandise, team-building activities, and massage certificates.
Categories
About Social Discovery Group
Social Discovery Group builds and operates consumer social-discovery and entertainment platforms that help people meet and interact online, and it also invests in related IT startups. The company monetizes its products as consumer internet services and runs multiple brands in dating, live-streaming, and chat experiences. Founded in 1991 and headquartered in Malta, it is privately held and works with distributed teams serving users in multiple countries.