• Design multi-agent architectures with robust state management, memory, and routing. • Choose and implement leading frameworks such as LangGraph/LangChain Agents, Microsoft AutoGen, CrewAI, LlamaIndex Agents, Semantic Kernel, or Haystack Agents—and justify trade-offs. • Build modular components (planners, tool registries, policy guards, evaluators) that are reusable across clients and domains. • Integrate enterprise tools and data sources via function/tool calling, webhooks, and event-driven flows (Queues/Service Bus/Functions). • Implement retrieval-augmented generation (RAG) patterns with vector stores (Azure AI Search, pgvector, MongoDB Atlas, Pinecone, Weaviate, Milvus) and structured knowledge (SQL/Graph). • Add deterministic fallbacks, circuit breakers, and caching to keep latency and cost predictable. • Define SLIs/SLOs for agent runs; implement tracing, metrics, and logging (e.g., Langfuse + OpenTelemetry) and build dashboards for run-level analytics. • Create evaluation harnesses (automatic + human-in-the-loop) using tools such as Ragas, DeepEval, promptfoo to measure groundedness, task success, safety, and cost. • Productionize with CI/CD, environment promotion, feature flags, and canary strategies; optimize cost-per-task and time-to-success. • Enforce content and safety policies (redaction, classification, guardrails) with policy-as-code; implement role/tenant isolation and data minimization. • Collaborate with security teams to align to ISO 27001/SOC 2/NIST/HIPAA/GDPR contexts; deliver audit-ready evidence for agentic workflows. • Build privacy-first patterns (no data exfiltration by default, least-privilege tool access, secure prompt/trace storage). • Work directly with enterprise client teams to translate business processes into agentic designs; present trade-offs and proofs-of-value that lead to production. • Partner with solution leads to create domain-specific agents (e.g., RFP assist, incident RCA drafting, knowledge ops) and reusable templates.

• Design and implement AI pipelines for detection, reasoning, prioritization, and decision support • Build agent-based and workflow-driven AI systems operating over security data • Work with structured and unstructured data across logs, alerts, telemetry, and external intelligence • Collaborate on infrastructure, deployment, and scaling of AI services • Implement monitoring, evaluation metrics, and guardrails • Work closely with security domain experts to encode real-world reasoning into AI systems

• Optimize model inference using advanced techniques including quantization (GPTQ, AWQ, GGUF), distillation, pruning, and speculative decoding • Build and integrate GenAI capabilities beyond LLMs, including computer vision, image generation (Stable Diffusion, FLUX), and multimodal models • Design and implement pre-processing and post-processing pipelines, including prompt engineering, structured output parsing, guardrails, and context management • Build RAG systems, embedding pipelines, and semantic retrieval architectures for enterprise AI applications • Drive model selection, benchmarking, and cost/performance trade-off decisions across AI services • Build evaluation frameworks to measure model quality, latency, reliability, and production performance • Build production AI systems that go beyond experimentation and notebooks, focusing on scalability, reliability, and maintainability • Collaborate closely with platform, infrastructure, and product teams to deliver integrated AI services • Contribute to AI platform architecture and long-term technical direction • Participate in the full lifecycle of AI systems, from research and prototyping to production deployment and operations

• You will build the systems that make Alaro’s product meaningfully useful to lawyers: agents, extraction tools, and controllers that automate repetitive work and reliably execute delegated tasks. • At a high level: we are building a Cursor-like experience for legal documents and legal research - where an agent can search, reason over sources, and propose or apply edits while the lawyer stays in control. • This is an engineering-first applied AI role. To succeed, you’ll need to understand the full product surface (editor, retrieval, dashboard workflows), so you can spot where an agent should be given control - and ship it end-to-end. • Document editing: an agent proposes changes and generates/applies patches to DOCX documents, built for legal reality: multiple collaborators, tracked changes, strict formatting, and high expectations for correctness. • Search & retrieval: when a lawyer asks “search legislation” or “find similar contracts in our database,” the agent decomposes that into targeted queries, retrieves the right sources, and grounds outputs with citations. • Deep research workflows: solving legal problems requires iterative investigation, where agents make progress step by step while keeping their active context focused and manageable. • Training models: fine-tuning models becomes a necessity to optimize the above workflows. As internal data matures, the agent can learn from past interactions to become more intelligent.