LMTS - Lead AI/ML Engineer (LLM Post-Training)
Salesforce3 days ago
Bengaluru, India or Hyderābād, IndiaStaff+
Responsibilities
- Train, fine-tune, adapt, evaluate, and serve open-weight language models for enterprise workflows.
- Own post-training experiments including SFT, LoRA/QLoRA, DPO, ORPO, GRPO, RLVR, distillation, and other preference or reinforcement-learning methods.
- Generate, curate, filter, deduplicate, version, and quality-review synthetic instructions, conversations, reasoning traces, tool-use trajectories, structured outputs, counterexamples, and preference data.
- Build the data engine for provenance, contamination controls, dataset splits, packing, tokenization, and loss masking.
- Design held-out evaluations, human and model reviews, error taxonomies, ablation studies, regression analysis, safety checks, and quality metrics.
- Investigate failures across data, objectives, model capacity, decoding, context handling, tools, prompts, retrieval, and serving behavior.
- Build production-grade model-serving, inference, enterprise-system, tool-use, structured-output, function-calling, and multi-step agent capabilities.
- Improve observability, traceability, testing, developer workflows, scalability, security, reproducibility, performance, and operational reliability.
- Collaborate with domain experts to create independently verifiable training and evaluation material while protecting sensitive data and preserving provenance.
- Lead architecture and design decisions, review model and application changes, mentor engineers, and provide technical guidance across the team.
Requirements
- 10+ years of software engineering experience, including technical leadership on production systems.
- Strong Python engineering experience building and maintaining reliable production systems.
- Hands-on ownership of LLM post-training for open-weight language models using SFT and LoRA/QLoRA; API integration alone does not qualify.
- Hands-on experience generating, filtering, reviewing, and versioning task-specific training datasets, including synthetic instructions, conversations, reasoning traces, tool calls, preference data, or structured outputs.
- Strong practical knowledge of chat templates, tokenizer behavior, sequence packing, loss masking, optimization schedules, checkpointing, distributed or memory-efficient training, and inference tradeoffs.
- Experience creating and evaluating reasoning, tool-use, or structured-output training data and distinguishing correct outcomes from plausible but low-quality traces.
- Experience with held-out datasets, quality metrics, experiment design, ablation studies, regression analysis, error taxonomies, and human-review workflows.
- Strong software-engineering fundamentals including API design, testing, debugging, observability, version control, and technical writing.
- Ability to investigate ambiguous failures across model behavior, data, prompts, retrieval, tools, and backend services.
- Demonstrated leadership in architecture, design reviews, code reviews, and mentoring engineers.
- Ability to work across application and ML layers, use AI tools in development workflows, and apply advanced prompt engineering to reliable and secure production systems.
- Preferred experience includes enterprise platforms and APIs, complex enterprise or regulated workflows, preference optimization or reinforcement learning, distributed or resource-constrained training and inference, retrieval evaluation, security and data governance, and measured production AI releases.
- Candidates should be able to provide verifiable examples of personally owned model-training projects, datasets, evaluation methods, production incidents, and technical artifacts.
- A PhD, publications, or prior domain experience is not required.
Benefits
- Salesforce provides benefits and resources intended to support balance and well-being.
- The role follows Salesforce’s hybrid work model.
- Salesforce provides reasonable accommodations during the application and recruiting process.
Categories
About Salesforce
Salesforce builds cloud-based customer relationship management software and a broader Customer 360 platform for sales, service, marketing, commerce, and analytics, sold by subscription to businesses and public-sector organizations. Founded in 1999 and headquartered in San Francisco, it trades on the NYSE under the symbol CRM. Its portfolio includes Sales Cloud, Service Cloud, Marketing Cloud, MuleSoft integration, Tableau analytics, and Slack for collaboration, with extensive developer tools and APIs.