Rapid7

Principal AI Engineer

Rapid7
Apply
3 days ago
Pune, IndiaStaff+

Responsibilities

  • Define architecture and technical roadmaps for LLM and agentic AI systems across Detection and Response and Exposure and Remediation.
  • Set model strategy, including when to use frontier or open-source models and when to fine-tune, distill, or post-train them.
  • Fine-tune, distill, and post-train open-source LLMs on security data through data curation, training, evaluation, and deployment.
  • Establish engineering standards for context engineering, inference optimization, and evaluation across teams.
  • Optimize LLM inference at scale using techniques including speculative decoding, KV-cache management, continuous batching, quantization, and custom kernels.
  • Establish evaluation systems as release gates for AI systems.
  • Lead design reviews, resolve complex technical trade-offs, prototype and build critical model, training, and inference code, and mentor Staff and Senior engineers.
  • Shape product and platform investments and align Engineering, Product, and other teams around shared technical direction.

Requirements

  • 12+ years of software or ML engineering experience, including 2+ years building production LLM or agentic AI systems.
  • Track record of building scalable, production-grade AI systems used across multiple teams or products, with organization-wide technical leadership and mentoring impact.
  • Deep understanding of transformer architecture, attention, tokenization, positional encodings, model training, and model serving, including the ability to implement core components from scratch.
  • Expert-level PyTorch experience with custom model code, training loops, mixed precision, and distributed training such as FSDP and DeepSpeed.
  • Hands-on experience with full and parameter-efficient fine-tuning, LoRA, QLoRA, distillation, and preference- or reinforcement-learning-based methods such as DPO and GRPO.
  • Experience with retrieval, memory, and context management for long-running agent workflows.
  • Deep knowledge of prefill and decode, KV-cache management, PagedAttention, prefix caching, continuous batching, speculative decoding, FlashAttention, quantization, and serving at scale with vLLM, SGLang, or TensorRT-LLM.
  • Experience defining evaluation strategies and benchmarks for LLM and agent systems.
  • Strong predictive AI, classical ML, distributed systems, and Python fundamentals, with judgment to select approaches based on accuracy, latency, and cost.
  • Strong communication, cross-functional collaboration, customer focus, and ability to explain technical trade-offs to engineering, product, and leadership audiences.
  • Preferred experience includes pre-training LLMs from scratch, writing GPU kernels with CUDA or Triton, profiling GPU workloads, security-domain work, running AI workloads on AWS at scale, and AI publications, patents, or open-source contributions.

Categories

AI ResearchML Engineering
Rapid7

About Rapid7

5,001-10,000 employees

Rapid7 builds a cloud security operations platform and services for enterprises, combining vulnerability management, detection and response (SIEM/XDR and MDR), application security testing, and security automation. Its Insight platform products include InsightVM, InsightIDR, InsightAppSec, and InsightConnect, sold as subscriptions with optional managed services. Founded in 2000 and headquartered in Boston, the NASDAQ-listed company (RPD) serves more than 11,500 customers worldwide.

Contact me