Note: The job is a remote job and is open to candidates in USA. Ditto builds systems that enable resilient, real-time data synchronization at the edge. The AI Engineer will build and operate the Symphony agent execution platform, including agent runtimes, typed tool interfaces, evaluation systems, model infrastructure, telemetry, and developer tooling. The role partners with internal engineering teams to make agentic software development reliable, measurable, and scalable.
Responsibilities
- Build the agent execution runtime: Design and ship the layer that compiles a brief into a running team of agents — topology and handoffs, deterministic gates, the shared message bus, session lifecycle, and the recovery behavior that keeps a long run from losing its work
- Design the tool surface agents work through: Replace "here is a shell, please be careful" with typed, versioned toolpacks backed by reviewed repository recipes — sandboxed execution boundaries, lease handling, process-tree ownership, and MCP servers that expose our internal systems to agents safely
- Make quality measurable, then improve it: Build the evals, benchmarks, and metric sets that score a run; instrument the system with per-phase timing, structured error taxonomies, and cost attribution; and use that signal to tune prompts, context, and topology rather than guessing
- Own the model substrate as a swappable input: Bind briefs to capability classes instead of vendor model names, work with routing and inference infrastructure across hosted and on-prem serving, and tune context, caching, and model selection against real cost, latency, and accuracy data
- Partner with service teams and raise the bar: Work directly with Ditto's engineering teams to turn their workflows into briefs they trust, establish the patterns others compose from, and mentor engineers as agentic development becomes the default path rather than the experiment
Skills
- 5+ years of professional software engineering experience, including meaningful time building and operating LLM-backed systems in production — not prototypes or demos, but something real people depended on
- Hands-on depth with agentic systems: tool and function calling, multi-agent orchestration, context and prompt engineering, structured output, and a practical understanding of where these systems fail and why
- Strong developer-tooling instincts from building the things engineers depend on — CLIs, build systems, CI/CD, internal platforms — with the empathy for the daily loop that comes from having been the person on call for it
- A rigorous approach to evaluation. You have built evals or benchmarks, and you reach for a measurement before an opinion when arguing that one approach beats another
- Strong Python (async, typed interfaces, Pydantic or similar) plus the systems fundamentals to work across a polyglot codebase — processes, sandboxing, filesystems, and the ability to debug something you did not write
- Demonstrated ownership and strong communication, with a track record of navigating ambiguity, setting technical direction, and working effectively with internal customers in a distributed, remote-friendly team
- Experience with MCP — building servers or clients, and designing tool interfaces that constrain what a model can do rather than trusting it not to
- Sandboxing and host isolation experience: process supervision, network and filesystem boundaries, resource leases, and the operational reality of macOS toolchains, Xcode, and iOS simulators
- Release engineering background — bundling, artifact provenance, benchmarking against a baseline, staged promotion, and rollback you have actually rehearsed
- Observability and cost accounting at scale, particularly OpenTelemetry, and experience making a spend number trustworthy enough to enforce budgets against
- Inference infrastructure experience: model routing and serving (vLLM or similar), prefix caching behavior, GPU capacity planning, or running open-weight models on owned hardware
- Fine-tuning or post-training experience with open-weight models, and a view on when it is worth the cost
- Familiarity with Rust, Swift, or Kotlin — enough to build tooling that has to understand the repositories it operates on
- A track record in fast-growing or startup environments, including 0→1 platform work where adoption had to be earned team by team rather than mandated
Benefits
- Meaningful equity
- In the US, health, dental, vision, life, and disability insurance
- In the US, a 401(k)
- In the US, flexible spending accounts
- Flexible time off
- Remote work
- Access to the Atlanta office as a place to work or meet with teammates
Company Overview
Company H1B Sponsorship