Job Specifications
About Persistent
We are an AI-led, platform-driven Digital Engineering and Enterprise Modernization partner, combining deep technical expertise and industry experience to help our clients anticipate what’s next. Our offerings and proven solutions create a unique competitive advantage for our clients by giving them the power to see beyond and rise above. We work with many industry-leading organizations across the world, including 20 Fortune 50 companies and 4 of the 5 top banks in both the US and India, and numerous innovators across the healthcare ecosystem.
Our disruptor’s mindset, commitment to client success, and agility to thrive in the dynamic environment have enabled us to sustain our growth momentum. Persistent has been recognized across top industry platforms for innovation, leadership, and inclusion. We have delivered 23 sequential quarters of growth with $422.5M in Q3 FY26 revenue, up 4.0% Q-o-Q and 17.3% Y-o-Y growth. Our 26,500+ global team members, located in 18 countries, have been instrumental in helping the market leaders transform their industries. We won the 2025 ISG Star of Excellence™ Award for AI and Data Excellence and were named a Leader in the Everest Group Talent Readiness for Next-generation Data, Analytics and AI Services PEAK Matrix® Assessment 2025.
About Position
Role:AI Architect
Location: Toronto Canada
Experience: 12+ years
Job Type: Full Time / Contract
Mandatory Skills: AI, ML, Agentic AI, LLM,
We are seeking an AI Architect to lead the design of next-generation enterprise AI systems centered on Agentic AI, advanced RAG patterns, multimodal model integration, and secure orchestration frameworks. This role focuses entirely on model integration, inference optimization, safety, orchestration, document ingestion, evaluation, and architecture. The architect will drive standards for multi-agent orchestration, tool calling, governance, and design patterns across business and engineering teams.
Expertise You'll Bring
AI & ML
Agentic AI
RAG / Graph RAG
LLMs, Vision-LLMs, Diffusion Models
Inference optimization (vLLM, TGI)
Systems & Architecture
Multi-agent orchestration
Tool calling & safe action schemas
Secure API integration
End‑to‑end AI application design
Data & Platform
Databricks ML, Unity Catalog, MLflow
Vector Search & retrieval systems
Document ingestion & OCR technologies
AI Safety & Governance
Guardrail frameworks
Evaluation tools
Responsible AI compliance
What You’ll Do
AI Strategy, Architecture & Governance
The AI Architect will define enterprise AI architecture that incorporates Agentic RAG, Graph RAG, Master-Agent orchestration, A2A agent communication, and MCP-based extensibility with tool calling. This involves shaping standards for how agents plan, reason, ground their responses, call tools safely, and collaborate autonomously. The role establishes architectural guardrails for tool schemas, action validation, lineage tracking, and auditability. It also sets direction for LLMOps and AgentOps patterns, ensuring alignment with data governance, responsible AI frameworks, and enterprise security requirements.
AI Solution Architecture
This role leads the end-to-end architecture of AI applications such as intelligent copilots, autonomous agents, workflow orchestration systems, knowledge-grounded assistants, and agentic marketing solutions. The architect integrates open-source and commercial LLMs, Vision-LLMs (e.g., Florence‑2, LLaVA, Qwen-VL), and diffusion models for text‑to‑image and emerging text‑to‑video capabilities. Rather than training models, the architect ensures secure, optimized inference and efficient orchestration using frameworks like vLLM, TGI, Hugging Face pipelines, and Databricks-first pipelines for governance (Unity Catalog), lineage tracking (MLflow), and semantic retrieval (Databricks Vector Search). Tool calling plays a central role as the architectural pattern for enabling LLMs and agents to safely interact with enterprise systems, APIs, knowledge stores, and other agents.
Knowledge Base Ingestion (PPT/PDF) Pipeline Architecture
The architect designs standardized, scalable ingestion workflows for enterprise PPT/PDF content using tools such as Unstructured.io, MarkItDown, Apache Tika, PyMuPDF, and python‑pptx. These pipelines incorporate OCR and layout understanding through LayoutParser, docTR, PaddleOCR, and OpenCV, as well as Vision-LLMs for interpreting charts, diagrams, and complex layouts. Ingestion workflows support semantic chunking, metadata enrichment, table extraction, media handling, and slide-aware structuring, all governed through Unity Catalog with indexing and retrieval powered by Databricks Vector Search. Tool calling is incorporated to allow agents to retrieve, interpret, or update knowledge assets securely and dynamically.
AI Safety, Guardrails & Evaluation
The AI Architect leads the design of safety, grounding, and evaluation systems using frameworks like DeepTeam and NVIDIA NeMo Guardrails to enforce policy