About this role
You will own the agent harness layer end-to-end, building and maintaining the software that runs agents against files, tools, and documents with permissions, memory, and unattended execution. The team at EnCharge AI builds a novel in-memory-computing architecture that delivers a 10x step-function improvement in compute energy efficiency and performance for AI inference workloads. Working closely with researchers and engineers, you will design session semantics, enforced permissions, human-editable memory files, orchestrators, and reliable output systems that adapt as open-weight models and bespoke checkpoints frequently change.
This role exists to ensure the infrastructure surrounding the hardware can successfully serve evolving models with robust tool-calling, security reviews, and evaluation suites. Ideal candidates bring extensive backend or ML infrastructure backgrounds, a track record of running production agent loops, and the product instincts needed to thrive in ambiguous, fast-moving environments.
What you will do
- Design and maintain the entire agent harness architecture, covering loops, safe execution, context management, and orchestration.
- Develop custom components without open-source equivalents, including session semantics, enforced permissions, and human-editable memory systems.
- Adapt the infrastructure to keep pace with changing models by updating prompt formats, tool-call schemas, stop conditions, and evaluation pipelines.
- Implement validation, repair, retries, and fallback mechanisms to ensure reliable tool use across models with uneven capabilities.
- Build internal and customer-facing agents, tools, and MCP servers while conducting security reviews before deployment.
- Create comprehensive evaluation harnesses to monitor regression, cost, latency, and quality on every model or harness change.
- Establish clear interfaces such as session APIs, command-line interfaces, graphical user interfaces, and endpoints for external tools.
What you need
- Bring over 10 years of professional software engineering experience focused on backend systems or machine learning infrastructure.
- Demonstrate strong proficiency in Python alongside at least one systems programming language like Go, Rust, or C++.
- Have successfully shipped and operated production agent loops that feature multi-step workflows, tool use, and unattended execution.
- Possess hands-on experience working with retrieval-augmented generation, context management, and memory implementations for large language model applications.
- Understand sandboxing, isolation, and permission models designed for automated software systems.
- Have direct experience running open-weight models with a solid grasp of how quantization and serving choices affect model behavior.
- Thrive in fast-paced and ambiguous environments where you independently define the product roadmap.
Nice to have
- Contributions to open-source agent harnesses or coding agents
- Experience with agent benchmarks like SWE-bench or Terminal-Bench and building internal task suites
- MoE serving familiarity such as expert placement, tensor parallelism, and quantization
- Observability for large language model systems
- Document parsing, indexing pipelines, and desktop or GUI application experience
Skills and tech stack
Questions candidates ask
Where is this position located?
This role is based in Bengaluru, Karnataka, with remote-friendly options throughout India and occasional travel.
What will I actually be working on?
You will own the harness layer that runs agents against files, tools, and documents with permissions, memory, unattended execution, and real outputs. This involves building custom components, managing context, and ensuring reliable tool use across frequently changing models.
What technical background is required?
Candidates need over 10 years of software engineering experience in backend systems or machine learning infrastructure, strong Python and systems language skills, and production experience with agent loops, RAG, and sandboxing.