ml engineer for dialogue systems

Social Discovery Group is a group of social discovery companies that addresses loneliness, isolation, and disconnection by transforming virtual intimacy into a new normal. Its portfolio includes social entertainment platforms connecting people online across cultures and regions.

• Own the ML strategy for dialogue systems, prioritize development, and connect initiatives to business metrics such as ARPU, retention, and chat depth;
• Lead a team of 3 ML engineers, set the technical bar, distribute work, hire and let go, and develop retained employees;
• Drive LLM post-training and model adaptation end to end, including SFT, LoRA/QLoRA, preference optimization, and dataset construction;
• Design agent harnesses and tool-using LLM systems with tool calling, structured outputs, routing, retries, memory, context, and guardrails;
• Build the evaluation layer with offline and pairwise evaluations, LLM-as-a-judge, and regression suites correlated with A/B outcomes;
• Reduce dialogue failure modes such as loops, contradictions, persona drift, context loss, and generic replies;
• Keep inference efficient in terms of latency and cost per message;
• Stay hands-on for approximately 10–20% of the time through prototyping, agent-trace debugging, and reviewing the team’s work.

• Hold a technical degree and have a hands-on ML engineering background with personally trained and shipped models;
• Have experience leading a team of 3–5 ML engineers, including hiring and performance decisions;
• Demonstrate expert-level Python knowledge;
• Understand transformer architecture and modern LLM behavior;
• Have hands-on experience with training, fine-tuning, and evaluation;
• Have production experience with LLM systems, including APIs, streaming, batching, fallbacks, cost/latency trade-offs, and observability over traces and transcripts;
• Be able to read current research and turn it into a prototype, an evaluation, and a shipped change with measurable impact;
• Be fluent in Russian;
• Be ready to work in CET (±2) hours;
• Nice to have: experience beyond text, including computer vision, image generation, and multimodal systems; long-running conversations and character consistency; vLLM / TGI / SGLang; DeepSpeed / FSDP / Accelerate; quantization; safety classifiers.

• Remote opportunity;
• Vacation: 28 calendar days per year;
• 7 Wellness days per year;
• Bonuses up to $5000 for recommending successful applicants;
• 50% Payment for professional training, international conferences, and meetings;
• Corporate discount for English lessons;
• Health benefits, including compensation of up to $1,000 gross per year per employee for eligible employees without corporate medical insurance;
• Equipped workplace and necessary equipment in company offices or co-working locations;
• Reimbursement of workplace costs up to $1000 gross once every 3 years in other locations;
• Internal gamified gratitude system with bonuses exchangeable for merchandise, team-building activities, and massage certificates.