Senior AI Engineer — Agent Platform & Document Intelligence

About SmartAdvocate

SmartAdvocate is an industry-leading legal case management platform used by 20,000+ legal professionals across North America, from solo practitioners to large multi-office firms. We are embedding advanced AI directly into the core of the platform to automate complex legal workflows, modernize firm operations, and generate deep document intelligence.

What you would be joining

This is not a greenfield AI experiment. SmartAdvocate already runs a suite of AI services in production: document classification and data extraction, medical chronology generation, demand letter drafting, case chat over vectorized case documents, an agent runtime, and an MCP server that lets external AI platforms operate directly against case data.

Your work is the next layer: an agent platform that runs case work autonomously in the background, and document intelligence that lets a generated legal document point back to the exact line of the exact record it came from.

Position overview

We are looking for a hands-on senior engineer to build two things that reinforce each other: background AI agents that do real case work inside our platform, and generative document engines whose every statement can be traced to its source.

This is a build-and-operate role. You will design these systems, ship them, and own them in production.

Key responsibilities

Autonomous and event-driven agents

• Architect background AI agents capable of reasoning, decision-making and multi-step workflow execution inside the case management system.
• Build document intelligence pipelines that extract, summarize, cross-reference and act on unstructured legal documents — medical records, pleadings, discovery, court orders and correspondence.
• Build trigger-based systems where agents watch case status changes, new documents and incoming data, and fire the right downstream action.
• Design the human side of autonomy: work assignable to an agent, a review queue where a supervisor approves or corrects it, escalation when the agent does not know what to do, and feedback that measurably improves the agent next time.
• Establish guardrails — permission scoping, approval gates on consequential actions such as sending mail or filing with a court, fallbacks, structured logging and complete auditability.

Generative drafting and visual source attribution

• Build generative synthesis pipelines producing structured legal artifacts: chronological medical summaries, demand letters, case evaluations and discovery drafts.
• Implement exact citation linkage that maps generated text back to source document coordinates — page, paragraph and bounding box.
• Develop interactive viewer capabilities, or the API payloads behind them, that render hyperlinks inside generated drafts, jump the reader to the precise location in the source PDF, and visually highlight the cited text with its surrounding context.
• Write extracted data back into structured case fields with visible provenance: what the AI entered, when, and whether a human has confirmed it.

Retrieval and quality

• Build RAG over large, messy, multi-page document sets: chunking strategy, hybrid retrieval, reranking, metadata filtering, and version-aware retrieval so an answer is never drawn from a superseded record.
• Treat evaluation as a first-class deliverable — golden datasets, faithfulness and groundedness scoring, hallucination detection, and regression suites that run in CI as a promotion gate before anything reaches a law firm.

Platform and integration

• Design clean REST APIs and event-driven architecture bridging AI services with the core application.
• Extend our MCP (Model Context Protocol) server so external AI platforms can operate against case data safely: tool design, scoping, authentication and rate control.
• Handle multi-tenant realities: per-tenant isolation, usage metering, quotas, token and cost accounting, and model tiering to keep unit economics sane.
• Own these services in production: deployment, monitoring, alerting and incident response.

Required qualifications

• Experience with Legal software development (case management, intake as examples)
• Production LLM application engineering — systems you shipped and still operate, not prototypes. OpenAI and Anthropic APIs, structured outputs, function calling, prompt and context engineering, cost and latency tuning.
• Agentic workflows built with orchestration frameworks or custom agent loops — LangChain, LangGraph, LlamaIndex, AutoGen, OpenAI Agents SDK or equivalent — including tool use, state, memory, retries and human-in-the-loop patterns.
• Retrieval-augmented generation at document scale: embeddings, vector databases, chunking, hybrid search, reranking, and grounding answers with citations.
• Document AI: multi-modal PDF parsing, OCR pipelines, layout analysis, and visual grounding or bounding-box extraction on complex, scanned and low-quality documents.
• AI evaluation and quality engineering: golden sets, faithfulness scoring, hallucination detection, and quality gates wired into CI/CD.
• Strong Python microservices (FastAPI/async) and clean REST API design.
• Event-driven systems: queues, event brokers and asynchronous execution patterns — RabbitMQ, Kafka, Azure Service Bus or background workers.
• Production ownership: Docker, CI/CD, monitoring, structured logging and tracing, and having personally debugged a live outage.

Nice to have

• MCP (Model Context Protocol) — having built or operated an MCP server, not just consumed one.
• Multi-tenant SaaS platform engineering: RBAC, audit logging, usage metering, quotas and per-tenant cost accounting.
• C#, .NET / ASP.NET and Microsoft SQL Server — our core platform, though the AI services are Python.
• PDF rendering tools and viewers (PDF.js, Syncfusion, PSPDFKit) and dynamic coordinate-based highlighting.
• Prior experience developing software for the legal, medical, insurance or other regulated industries.
• Understanding of HIPAA, PHI handling and attorney-client privilege constraints for AI applications.
• Fine-tuning, distillation or model routing to reduce inference cost.
• Handwriting recognition and vision-model document reading.

How we hire

We read work, not keywords. Our process is deliberately short:

• Resume and portfolio review
• A 45-minute conversation about systems you have built and operated
• A short technical exercise against a realistic problem from our actual domain — not a whiteboard puzzle
• A final conversation with our CTO and the AI team

What we offer

• SALARY RANGE — $120,000.00 - $160,000.00 per year.
• Comprehensive benefits package
• LOCATION: Remote (US) — or Hybrid, Melville NY
• Reports to the CTO; relationship to the existing AI Architect and AI development team
• The chance to lead high-impact AI work on an established platform with real customers, real data, and problems that are not yet solved anywhere in legal tech

Pay: $120,000.00 - $160,000.00 per year

Benefits

• 401(k)
• 401(k) matching
• Dental insurance
• Flexible spending account
• Health insurance
• Health savings account
• Life insurance
• Paid time off
• Parental leave
• Vision insurance

Experience

* Legal software development: 1 year (Required)

Work Location: Remote