Founding AI Engineer

Harden; San Francisco (CA); Hybrid; USD 120,000 - 250,000; Full time

Job summary

Harden is seeking a founding AI Engineer to design and ship RL-driven optimization loops for security-hardening of AI-generated apps. You’ll work across research, ML engineering, and security engineering to production systems inside CI/CD and BYOC environments.

You will build backend services, train reward models, and scale distributed training for AI-assisted code analysis. A hands-on, collaborative engineer with strong Python and systems skills will thrive in our fast-moving startup.

Qualifications

• Proficient in Python; TypeScript/Go/Rust bonus for systems.
• Strong LLM systems intuition with tool-use and telemetry.
• Have shipped production ML or security-enabled systems.
• Experience designing experiments and evals for ML models.
• Fluent data pipelines, logging, and reproducible workflows.
• Excellent communication in fast, ambiguous environments.
• 2+ years building ML systems.
• Familiar with authentication, authorization, secrets, sandboxing, prompt injection.

Responsibilities

• Design RL loops for multi-step, tool-using agents for security hardening.
• Build backend services for training, evals, and policy updates.
• Train reward models; run A/B tests with security signals.
• Scale distributed training/serving for AI-assisted code analysis and deployments.
• Turn research papers into maintainable code in Harden.
• Collaborate with product and customers to define measurable rewards.
• Combine LLM reasoning with deterministic SAST/SCA and runtime telemetry.

Skills

Python

TypeScript

Go

Rust

LLM systems

Production systems

Experimentation

Data pipelines

Communication

ML systems

Security concepts

Job description

# Founding AI EngineerRemote (US) or Hybrid (SF Bay Area)Full-time$120,000–$250,000 + equity## About UsHarden is the enterprise security and governance platform for AI-built apps. Every AI-generated application ships with the same gaps — hardcoded secrets, missing auth, uncontrolled egress, no audit trail. Harden closes them automatically: build-time SAST and SCA scanning with auto-remediation, a one-page approval report that gates CI/CD, and runtime enforcement that covers egress controls, prompt-injection defense, AI cost caps, and tamper-evident audit logs — all deployed inside the customer's own cloud.We're a venture-backed startup based in the SF Bay Area. The founding team combines AI/ML research leadership at Google DeepMind and Amazon with product leadership at Verkada and AWS. Enterprise customers are already in production and we're bringing on exceptional engineers to build the infrastructure that makes enterprise AI adoption safe and scalable.## About the RoleYou'll work across the stack to design, train, and deploy RL-driven optimization loops for AI agents. At Harden, those loops help inspect AI-generated applications, reason about security gaps, generate reliable fixes, and evaluate whether applications are ready for enterprise deployment. You need to be deeply technical, hands-on, and comfortable moving between research, ML engineering, security engineering, and production systems (APIs, infra, SLAs).This role sits at the intersection of LLM systems, security engineering, and production software. The goal is not to build demos; it is to ship AI-assisted hardening systems that enterprises can trust inside real CI/CD and BYOC environments.## What You'll Do* •Design RL loops for multi-step, tool-using agents (planning, retrieval, coordination), applied to security hardening and remediation workflows for AI-generated applications* •Build backend services for training, evals, and online policy updates, including services that rank findings, generate fixes, and produce Harden approval reports* •Train reward models from traces/preferences; run A/Bs & interleavings safely, with security-specific signals for remediation correctness, false positives, regression risk, and policy coverage* •Scale distributed training/serving for AI-assisted code analysis, generated-app evaluation, and customer deployment feedback loops* •Turn papers → running code → measurable uplift, then ship the useful pieces as maintainable Harden product capabilities* •Partner with product & customers: translate messy, real-world objectives into measurable rewards and robust policies for enterprise security, compliance, and runtime governance* •Combine LLM reasoning with deterministic SAST, SCA, SBOM, policy, and runtime telemetry signals to harden applications before they reach production## What We're Looking For* •Strong programming in Python; bonus for TypeScript/Go/Rust for systems and APIs* •LLM systems intuition including tool-use, planning, retrieval, structured outputs, and how evals/telemetry become learning signals* •You've shipped production systems and can debug/profile at speed* •Experimentation mindset: design clean evals, run ablations, read papers, and turn them into maintainable code* •Data & infra fluency: event/trace pipelines, schema design, reproducibility, and versioning for datasets, policies, and rewards* •Clear communication in a fast, ambiguous environment* •2+ years experience building ML systems* •Comfort working with security concepts such as authentication, authorization, secrets, dependency risk, sandboxing, prompt injection, and data exfiltration## Nice to Have* •RL proficiency with some real-world RL applications* •Experience with agent stacks: orchestration graphs, tool routers, retrievers, evaluation frameworks, and observability of traces* •LLM post-training experience: reward-modeling, preference data collection, safety/guardrail integration, structured evals* •Experience with security automation, code review agents, SAST/SCA tools, or vulnerability remediation