AI Software Engineer, Agent Harness
Salary not listed • India • Full-time
Posted 1mo ago
About the job
EnCharge AI is building the next generation AI platform, focusing on a novel in-memory-computing architecture for improved AI inference efficiency and performance. As AI demands evolve, we believe the underlying infrastructure must also advance. We are an experienced team of AI researchers, silicon & systems engineers, and architects, backed by leading investors, aiming to be the essential platform for the next wave of AI innovation.
This role involves owning the "harness" layer that enables agents to interact with files, tools, documents, and memory, managing permissions and unattended execution. This layer will be built using a combination of open-source and bespoke code, requiring you to keep pace with rapidly changing models and ensure reliable tool use across different model qualities.
Responsibilities
- Own the end-to-end harness architecture, including agent loop, safe execution, context management, knowledge base, memory, permissions, orchestration, outputs, interfaces, and observability.
- Build components for session semantics, enforced permissions, human-editable memory, orchestrator, and outputs, especially where no open-source equivalent exists.
- Adapt to evolving models by managing adapters, prompt formats, tool-call schemas, stop conditions, benchmarking, and evaluation.
- Ensure reliable tool use across models with varying tool-calling capabilities through validation, repair, retries, and fallbacks.
- Develop agents, tools, and MCP servers for internal and customer use cases, ensuring security reviews before deployment.
- Build the evaluation harness, including task suites, regression runs for every model or harness change, and tracking cost, latency, and quality per task.
- Define interfaces for session API, CLI, GUI, and an endpoint for existing tools.
Requirements
- 10+ years of software engineering experience in backend systems or ML infrastructure.
- Strong proficiency in Python and at least one systems language (e.g., Go, Rust, C++).
- Proven experience shipping and operating an agent loop in production, including tool use, multi-step workflows, and unattended runs.
- Hands-on experience with RAG, context management, and memory for LLM applications.
- Experience with sandboxing, isolation, and permission models for automated systems.
- Experience running open-weight models and understanding how quantization and serving choices affect model behavior.
- Comfort working in fast-moving, ambiguous environments with the ability to define roadmaps and strong product instincts.