Software Engineer, AI Agent Platform

About Chipforge
Chipforge is the platform that takes you from design intent to working, verified register-transfer level (RTL) hardware, starting with FPGAs and expanding to ASICs, cutting out the time, iteration cost, and tool complexity of doing it the traditional way.
Chipforge is the primary strategic subsidiary of Pathkey Limited (ASX: PKY), an Australian-listed applied-AI company, and operates across two entities: Chipforge Technologies Pty Ltd (Australia) and Chipforge Private Limited (Singapore).
About the role
This role builds the agent layer of the product: the system that takes a description of a design, produces HDL, checks that output against real verification tooling, and keeps working until the result holds up.
You will be working with primarily Python and FastAPI system running a multi-stage context pipeline. You would take technical ownership of the agents and set the direction for the module rather than picking up tickets around the edges. You will be working closely with other agents engineers.
This is a software engineering role. You will not be training models. You will be building production systems around one, in a domain where the answer is verifiable: the design either simulates or it
doesn’t.
What you’ll own
Build new agent capabilities - additional tool integrations, specialized agent roles, and new steps in the request lifecycle.
Extend the platform's agentic patterns as the field moves: better planning and delegation, memory, structured tool use, evolving provider APIs.
Optimize context management - what's included per turn, truncation and prioritization under token limits, measurable quality gains.
Tune cost and latency per request without degrading output.
Harden agentic loops - error handling, retries, fallbacks, graceful degradation.
Build evaluation harnesses that catch regressions before customers do.
Instrument tracing and logging so non-deterministic failures are debuggable.
Scale long-running requests - resume, durable state, multiple instances.
What you’ll need
4+ years of production Python, with FastAPI or relevant asynchronous frameworks. Agentic loops are streaming, concurrent and long-running by design.
Resilience engineering: retries and idempotency, backpressure, graceful degradation, reconnect-and-resume.
State design: you know how to get a stateful, streaming service running on more than one instance.
Production experience with agentic development: tool calling, structured output, streaming, retries, token budgeting, and error handlings for unexpected model outputs.
Comfort with SSE or WebSocket streaming and its failure modes: partial writes, dropped clients, reconnects that expect to resume
Good API design with clear boundaries. You can define a contract that other codebases build against and keep it stable while all of them keep changing.
Sufficient TypeScript knowledge to work on streaming, tool-execution in cross-repo changes.
PostgreSQL: schema design, migrations, and queries that stay fast as data grows.
Comfort operating with ambiguity. The architecture is still moving, and you will need to make design decisions without fixed playbooks.
Clear written communication. The team is distributed, and most design and review happens asynchronously.
Nice to have
Evaluation work on agents or code generation: building harnesses, defining pass criteria, measuring non-tangible performance, catching regressions.
Local and self-hosted inference (vLLM, SGLang, Ollama), and building a working API and system around self-hosting self-serving models.
Retrieval and ranking work, including RAG over code.
Good working knowledge of TypeScript and Go, able to do systems judgements across multiple layers and runtimes.
Exposure to hardware design workflows or EDA tooling such as Verilog simulation, linting or synthesis. You do not need a hardware background, but curiosity about the domain helps.
Experience building for air-gapped or on-premises deployment environments.
AWS architecture and infrastructure-as-code experience, including CI/CD pipelines and production observability (logging, tracing, alerting) for backend or AI inference systems.
Practical React and TypeScript capability to connect front-end interfaces to backend services or deliver complete features independently.
How we work
This is a remote-first role, open to candidates based in Australia or Singapore. We work closely as a distributed team across Australia, Singapore, and India, so comfort collaborating across time zones matters. Occasional travel for team or product sessions may come up, but the day-to-day is remote.

Work arrangement
No