Machine Learning Engineer - LLM & Agentic Code IntelligenceIntegrated Research Ltd. • United States
Machine Learning Engineer - LLM & Agentic Code Intelligence
Integrated Research Ltd.
- United States
- United States
About
Do you think in tokens, prompt graphs, and tool‑calling flows? At IR Labs you’ll be the founding Machine‑Learning Engineer for LLM & Agentic Code Intelligence, joining a lean squad of RL, graph, and compiler specialists to turn large language models into safety‑hardened developer copilots that ship to customers in weeks—not quarters. Your remit is genuinely end‑to‑end: design the agentic architecture, fine‑tune and distill code‑centric LLMs, orchestrate multi‑tool workflows, and watch your models deliver 10× impact for Fortune‑scale engineering teams. Because we run like a venture‑backed startup inside a public company, you’ll have both the autonomy to experiment rapidly and the resources to scale globally. If you thrive on deep technical puzzles, obsess over fast, clean code, and want your ideas in production fast, we want to meet you. What You’ll Do Architect and lead our agentic LLM stack—from research prototype to production MVP—balancing state-of-the-art techniques with real-world latency, security, and cost constraints. Design, fine-tune, distill, and serve code-focused language models using QLoRA, LoRA, AWQ, knowledge distillation, and RLHF/RLAIF to hit aggressive efficiency targets on GPU & CPU. Build secure multi-tool agents that orchestrate compilers, linters, search APIs, and internal knowledge graphs via function-calling frameworks (LangChain, LlamaIndex, Autogen) while enforcing strong guardrails and observability. Generate and curate high-quality synthetic datasets for tasks such as bug-fixing, refactoring, explanation, and test-generation; automate fuzzing, mutation, and self-instruct loops to enrich long-tail coverage. Implement active-learning loops that surface high-uncertainty code snippets, route them for human review, and automatically retrain models, closing the label-quality gap. Leverage low-level code representations (AST, CFG, LLVM/MLIR IR, byte-code) to build richer model inputs, evaluation metrics, and downstream reasoning graphs. Optimize inference & serving paths with TensorRT-LLM, vLLM, and Flash-Attention; deploy behind Triton, Ray Serve, or TGI with strict resource isolation and quota enforcement. Instrument, monitor, and defend the agent runtime—collect structured traces, detect prompt-injection, throttle abusive inputs, and provide detailed lineage for security audits (SOC 2/HIPAA/GDPR). Collaborate cross-functionally with compiler engineers, product, and DevSecOps to translate business needs into safe, reliable developer-facing features (auto-PRs, vulnerability alerts, code-review copilots). Publish internal best practices & mentor peers on LLM evaluation, safe prompt engineering, and cost-aware deployment strategies. Desired Skills and Experience
Qualifications 8 + years end-to-end ML experience, 5 + years in NLP/LLMs with a focus on code understanding or generation. Proven record shipping agentic systems that coordinate multiple tools/APIs in production; deep familiarity with OpenAI function-calling, LangChain Agents, or equivalent. Expert-level PyTorch (or JAX) plus Hugging Face Transformers, Accelerate, and PEFT; hands-on with Triton kernels, CUDA, or XLA a plus. Demonstrated success compressing and distilling foundational models (7 B → 1–3 B or 70 B → 8 B) while retaining task accuracy. Hands-on RL optimization (PPO/DPO/ReLoRA) for aligning models or agents with human preferences and latency budgets. Experience constructing and querying code corpora at 100 M + file scale; comfort with data lakes (Delta/Parquet), vector stores, and retrieval pipelines. Strong security mindset: OWASP-SAMM, prompt-injection defenses, jailbreak red-teaming, least-privilege IAM. Adept at observability (Prometheus, Grafana, OpenTelemetry) and cost tracking across GPU clusters. Clear communicator and mentor, able to translate complex LLM/agent trade-offs to stakeholders. Nice to Have’s Publications or OSS contributions in code-LLM distillation, agent tool orchestration, or compiler-aware ML. Experience with differentiable compilers (TVM, IREE) or code-intel graph frameworks (CodeQL, Joern). Background in privacy-preserving ML (DP-SGD, confidential compute) for corporate codebases. Familiarity with multimodal code + UI agent workflows (e.g., design-to-code generation). Our job descriptions often reflect our ideal candidate. If you have a strong foundation of relevant skills and a passion for this field, we encourage you to apply, even if you don't check every box. What We Offer Culture: Join a passionate, driven team that values collaboration, innovation, and having fun while making a difference. High‑Impact Ownership: Your code and ideas will go live in weeks, not quarters. Every engineer owns features end‑to‑end and sees their work land in production with Fortune‑grade customers. Innovation: Work on cutting-edge AI solutions that solve real-world problems and shape the future of technology. Growth: Opportunity for personal and professional growth as the company scales. Flexible Work Culture: Benefit from a flexible work environment that promotes work-life balance and remote work. Competitive Compensation: Receive a competitive salary, performance bonuses, equity participation and a generous benefits package. 401k with Employer Contributions Health Savings Account (HSA) Contributions with High Deductible Health Plan Short-Term/Long-Term Disability Insurance And more! Compensation Range $180,000 - $210,000 base $53,000 - $63,000 variable compensation Actual compensation offer to candidate may vary from posted hiring range based upon geographic location, work experience, education, and/or skill level. The pay ratio between base pay and target incentive (if applicable) will be finalized at the offer stage. At IR we celebrate, support, and thrive on difference for the benefit of our employees, our products, and our community. We are proud to be an Equal Employment Opportunity employer and encourage applications from all suitable candidates; we never discriminate based on race, religion, national origin, gender identity or expression, sexual orientation, age, or marital, veteran, or disability status.
#J-18808-Ljbffr
Languages
- English
Notice for Users
This job comes from a TieTalent partner platform. Click "Apply Now" to submit your application directly on their site.