AI Automation Engineer / LLM Workflow Engineer for Legal Document & Paralegal Automation

AI Automation Engineer / LLM Workflow Engineer for Legal Document & Paralegal Automation Overview We are a growing law firm looking for an experienced AI/software engineer to help us build an internal AI-powered paralegal system using Claude and potentially other LLM technologies. The goal is not to replace attorneys or produce unsupervised legal work. The goal is to create a reliable system that can take structured information already contained in our CRM/case-management system, combine it with our law firm’s templates, instructions, examples, and workflows, and generate high-quality first drafts for attorney review and final editing. We have many repetitive legal documents and forms where much of the information already exists in our CRM. Today, staff manually retrieve that information, select the correct template, populate the document, draft additional language, check it for completeness, and send it to an attorney for review. We want to automate as much of that workflow as reasonably possible. What We Want to Build The project will involve creating reusable Claude Skills, Claude Projects, prompts, workflows, scripts, API integrations, or other AI automation architecture necessary to accomplish tasks such as: 1. Retrieve relevant matter/client information from our CRM or other data sources. 2. Determine which information is needed for a particular legal form or document. 3. Map CRM fields and case information into the appropriate document fields. 4. Use our existing Word/PDF/document templates as the foundation for the output. 5. Apply document-specific drafting instructions and law-firm standards. 6. Generate narrative sections when information cannot simply be inserted into a field. 7. Identify missing information rather than hallucinating or guessing. 8. Flag inconsistencies or information requiring attorney review. 9. Produce a properly formatted draft document. 10. Route the draft to an attorney or staff member for review and final editing. Examples could include pleadings, discovery documents, client forms, correspondence, affidavits, financial documents, intake documents, checklists, and other repeatable legal work. A Typical Workflow Might Look Like Matter selected → CRM data retrieved → appropriate template loaded → required information identified → missing information flagged → AI drafting instructions applied → document generated → automated quality-control review → draft delivered for attorney review. We are looking for someone who can help determine the best technical architecture rather than simply following a specification we have already created. Important Part of the Project: Creating Reusable AI Skills We want to develop reusable Claude Skills or similar modular AI workflows for different legal tasks. For example, one skill might be responsible for creating a particular pleading. That skill could contain: * The firm’s approved template * Instructions for when the document should be used * CRM/data-field mappings * Required facts * Drafting rules * Formatting requirements * Examples of previously approved documents * Quality-control rules * Legal or procedural reference material * Instructions for handling missing or conflicting information * Rules identifying issues that require attorney review We envision eventually having a library of these modules so the AI behaves more like a trained paralegal following our firm’s established procedures. What We Need From You We need someone capable of working on both the AI/LLM side and the software/integration side of the project. Relevant experience may include: * Claude API / Anthropic API * Claude Skills or Claude Projects * OpenAI API or similar LLM platforms * Prompt engineering and context engineering * Agentic workflows * Structured outputs / JSON schemas * Function calling / tool use * REST APIs and webhooks * CRM integrations * Document automation * Microsoft Word/DOCX generation * PDF form filling or PDF generation * Template-based document generation * Python, TypeScript/JavaScript, or similar languages * Retrieval-augmented generation (RAG), where appropriate * Workflow tools such as n8n, Make, Zapier, or custom automation * Secure handling of confidential information * Logging, testing, versioning, and quality control for AI outputs Experience with Clio or other legal practice-management software would be particularly helpful. Legal-industry experience is a plus but not required if you have strong document-automation and LLM-workflow experience. Reliability Is Extremely Important This is legal work. We do not want an AI system that confidently invents information. The system should be designed so that it: * Never fabricates missing facts * Clearly identifies missing information * Preserves source information accurately * Distinguishes retrieved facts from AI-generated language * Flags conflicting information * Follows templates consistently * Produces reproducible outputs * Maintains an audit trail where appropriate * Includes attorney review as a required part of the workflow We are interested in engineering solutions that make LLM output more deterministic, testable, and reliable, rather than simply writing increasingly complicated prompts. Initial Project We would likely begin with one or two specific legal documents as a proof of concept. For the first workflow, we would provide: * Our existing template * Examples of completed documents * The information available within our CRM * Our drafting preferences * The rules our staff currently follows when preparing the document You would help us: 1. Analyze the existing manual workflow. 2. Determine the appropriate AI/software architecture. 3. Build the initial automation. 4. Create the necessary Claude Skill/project/instructions/prompts. 5. Connect the required data sources. 6. Generate the document. 7. Create validation and quality-control checks. 8. Test the workflow against multiple sample matters. 9. Document the system so additional workflows can be added later. If the proof of concept is successful, this could become a significant ongoing project involving many different legal documents and internal law-firm workflows. What We Are NOT Looking For We are not looking for someone whose only experience is writing ChatGPT prompts. We also are not looking for a developer who automatically assumes every problem requires a large custom software application. We want someone who understands both sides and can determine when the best solution is: * A Claude Skill * A well-designed prompt/workflow * An API integration * A small deterministic script * A document-generation engine * A database or structured-data layer * RAG * An automation platform * Or custom software The objective is to build the simplest reliable architecture that can ultimately scale. When Applying Please answer the following: 1. Describe an AI/LLM workflow you have built that took structured data and generated a document or other structured output. What was the architecture? 2. Have you worked directly with the Anthropic/Claude API, Claude Skills, Claude Projects, or Claude tool use? Please explain specifically what you built. 3. Have you integrated an LLM with a CRM, practice-management system, or other business database? Which systems? 4. What approach would you use to prevent the AI from inventing information when required information is missing from the CRM? 5. How would you separate deterministic document automation from portions of the workflow where an LLM is actually needed? 6. Have you generated Microsoft Word documents programmatically while preserving professional formatting? If so, describe your experience. 7. If we gave you a legal template, five examples of attorney-approved completed documents, and access to the corresponding CRM data, briefly describe how you would turn that into a repeatable AI workflow. Please include examples of relevant projects if available. Long-Term Opportunity Our larger goal is to create an internal AI system that functions like a highly trained drafting paralegal, capable of handling an increasing number of repeatable workflows while consistently following our firm’s templates, procedures, and drafting standards. Every substantive legal document will remain subject to attorney review and approval. For the right engineer, this could develop into a substantial long-term relationship as we automate additional processes throughout the firm.