AI Engineer

Part Time Paid

• Responsibilities Benefits: Flexible schedule Opportunity for advancement Training & development ABOUT THE ENGAGEMENT T12 Technologies LLC is a Maryland-based technology services firm delivering cloud, AI and automation, cybersecurity, contact center, and back-office modernization to federal, state, healthcare, and commercial clients. We are staffing a finance automation program for a commercial client: an accounts payable and vendor payment build on Oracle NetSuite, with an AI-enabled invoice extraction layer in Texas. You are the builder. Working to the architect’s design, you construct the Claude workflows that read invoices, extract and validate line-level data, apply GL coding rules, and route approvals - then tune them against live invoice volume during a parallel-run pilot until they hold up on real edge cases. This role also owns validation: there is no separate QA engineer on this program, so building the evaluation harness and proving accuracy against ground truth is part of the job. Scope of work Invoice automation build. Build Claude invoice intake, extraction and GL coding workflows, tuned against the client’s historical coding patterns and vendor mix. Confidence tiers. Configure auto-approve versus exception routing and implement human-in-the-loop thresholds for high-value transactions. Prompt engineering. Author, test and version the prompt library covering coding, duplicate rationale, exception summaries and anomaly analysis. Evaluation harness. Build ground-truth sets from historical invoices, accuracy scoring, and regression checks that run on every change. Validation evidence. Validate extraction and coding accuracy against agreed targets and document the results as acceptance evidence. Monitoring automation. Build Claude-assisted analysis and duplicate saved searches, producing investigation summaries and disposition drafts. Reporting workflow. Build a budget-versus-actual variance commentary workflow for the finance team. Go-live. Support cutover, capture baseline metrics, and participate in post-implementation hypercare. Required - references and work samples There is no universal certification for this role, so we verify through evidence of delivered work instead. References are contacted before any contract is issued. Two references minimum, three preferred, from production LLM work you personally built. Demos, pilots and proofs of concept do not qualify. For each reference: name, current title, organization, direct phone or email, the project you worked on together, and their role on it If you hold Anthropic, cloud provider, or ML certifications, list them with credential IDs. We verify them.

2+ years building production workflows on LLM APIs - Anthropic Claude strongly preferred - including prompt engineering, structured extraction, and confidence scoring. Strong Python and REST API integration; comfort working against a third-party ERP you did not build. Document extraction at production quality - invoices, forms, contracts, or similar semi-structured documents. Demonstrated experience building evaluation sets and regression suites for non-deterministic output; you can tell a genuine accuracy gain from noise. Structured output and schema validation with real handling for malformed or low-confidence results. Disciplined prompt iteration against real data rather than synthetic happy-path examples. Legal authorization to work in the United States as an independent contractor. Preferred SuiteScript 2.x or prior NetSuite / ERP integration experience. AP, finance, or accounting document processing background. OCR and document-capture pipelines. Experience on a team where an architect owns the design and you own the build.