Senior AI Orchestration Engineer

Senior AI Orchestration EngineerPython, FastAPI & Microsoft Agent Framework | FreelanceElchai Group | Hands-on backend development | Remote working by agreementBUILD THE WORKFLOWS BEHIND A GOVERNED AI OPERATING SYSTEMElchai is developing Governance Orchestration as one operational engine, with New Pulse as the authenticated human interface. The objective is to coordinate business workflows across separate customer, venture and function workspaces while preserving explicit human approval for consequential actions.We are seeking a senior freelance engineer to accelerate implementation of that engine's workflow packages. You will write working backend and orchestration code, connect it to approved platform interfaces, resolve package-level defects and hand over maintainable software.This is a development role. Our internal QA team provides independent validation. We need someone who builds the workflows, not someone hired only to test them.WHAT YOU WILL BUILDYou will extend an existing Python platform rather than introduce a parallel orchestration architecture. Assignments will come from a defined backlog and an internally approved specification.The work includes:

• Implementing function-level agents and multi-step workflows on Microsoft Agent Framework, including routing, execution steps, state transitions and structured outputs.
• Migrating approved business workflows into reusable function packages that use the existing shared core, content store and approval interfaces.
• Building scoped FastAPI services and provider adapters against frozen contracts, using synthetic data and fake providers during development.
• Implementing persistence and recovery behaviour through approved platform interfaces, including bounded retries, duplicate-event handling, timeouts and partial-failure reporting.
• Connecting workflow proposals and results to the existing approval and evidence interfaces without redefining their security rules.
• Delivering developer tests, small reviewable changes, defect corrections and documentation so our internal team can integrate and maintain the package.Potential assignments include inbound-response classification and follow-up drafting, research and grant-opportunity triage, marketing draft-generation workflows, client-success workflows and status reporting. These are examples of the programme backlog, not a promise to assign every area to one freelancer.YOUR FIRST ASSIGNMENTThe intended first engagement is one small, non-sensitive workflow build, not a test-only assignment or a rewrite of the whole platform.Before work starts, we will agree its scope, permitted files, input/output contracts, synthetic fixtures, expected behaviour, internal owner, reviewers, milestones and price. Our internal team will provide the approved company rules and technical handover rather than ask you to infer them.You will deliver a functioning package that runs from input to the agreed output or human-review handoff in the approved sandbox. The exact workflow and start date remain subject to scope, access and engagement approval. Additional packages may follow successful delivery and separate approval; continuation is not guaranteed.REQUIRED EXPERIENCEStrong Python backend engineering.You have personally built asynchronous services, APIs, database-backed applications and maintainable modules. FastAPI, PostgreSQL, SQLAlchemy and typed data validation are directly relevant.Hands-on orchestration delivery.You can show coded agent or workflow systems that handle state, tool calls, failures and recovery beyond a single successful demonstration. Direct Microsoft Agent Framework experience is strongly preferred. Comparable Python orchestration experience is relevant, but you must demonstrate the ability to work on our approved MAF baseline before assignment.Reliable execution.You understand how to prevent duplicate effects, distinguish a failed operation from an uncertain outcome, preserve approval requirements during retries and report incomplete results accurately.Contract and security discipline.You can implement workflow logic using existing identity, branch, approval and evidence interfaces without changing their authority model. You recognise when a dependency needs an internal decision.Delivery ownership.You write and debug the code, provide meaningful developer tests, explain your decisions, respond to review and leave another engineer able to run and maintain your work.Docker, Git-based collaboration and automated checks are expected. MCP/provider-adapter experience and OpenTelemetry/Langfuse familiarity are useful. Next.js/TypeScript is a bonus, not the main selection criterion.Experience limited to prompt writing, no-code automation, chatbot demos or QA-only assignments does not establish fit for this role.WHAT SUCCESSFUL DELIVERY LOOKS LIKEAn accepted package contains working implementation code, the agreed workflow behaviour, reproducible normal and failure-case tests, approved-interface compatibility, a concise defect record and a usable handover.Where the workflow has an approval step, it must wait for the appropriate human decision rather than treat model output as permission.Our internal QA and security reviewers validate independently. Architecture, central permissions, policy, approval mechanisms, shared migrations, credentials, real data, the New Pulse integration contract and release decisions remain internal. You implement the approved workflow logic that uses those services.ENGAGEMENT AND APPLICATIONWe intend to start with one hands-on freelancer on a paid, fixed-scope milestone after the required approvals.We are currently collecting applications and non-binding commercial proposals. Rate, milestone price, availability and time-zone overlap will be agreed before contracting.No unpaid custom work is required. Any practical build assessment will be separately authorised and paid.Send your application through this LinkedIn listing or the posting account with:
• One or two relevant systems you personally built, your contribution and existing evidence you are permitted to share.
• Your Microsoft Agent Framework experience, or the comparable orchestration framework you used.
• A concrete example of handling restart, retry, duplicate execution or a human-approval boundary.
• Availability, time zone and indicative rate or milestone-estimation basis.An anonymised technical account is welcome when client confidentiality prevents sharing code.Identify who would personally perform the work and disclose any proposed subcontracting.THE OUTCOME WE NEED IS WORKING ORCHESTRATION SOFTWARE THAT OUR INTERNAL TEAM CAN VALIDATE, INTEGRATE AND OWN.