AI Architect

Job Titl e : AI Architect

Employment Type : Contract

Location : Atlanta, GA / Charlotte, NC (Onsite)

Job Description :

Responsibilities:

Define platform and solution architecture for Azure Functions, Blob/Table Storage, Azure AI Foundry (Document Intelligence, Azure OpenAI), Azure AI Search, and Application Insights across environments (dev/test/prod).

Design the Agentic Registry domain model (agent definitions, tools, prompts, evaluation policies, routing) and align with enterprise IAM/Networking.

Own the access-model template that every new agent must satisfy (IAM scope, RBAC tier, PII handling classification) before it can enter the “Next” backlog tier, closing the “no access model defined” gap on agents like Cash Administration.

Establish observability (structured logs, traces, metrics, correlation ids), nightly/batch evaluation jobs, and quality KPIs for agents.

Oversee API contracts (FastAPI) and schema governance (Pydantic) for deterministic I/O with LLMs; champion prompt templating standards (Jinja2).

Lead security reviews (data residency, PII handling, prompt injection defenses, model safety filters, content policy); partner directly with CSO and Legal on security sign-off before any agent reaches production.

Own compliance alignment for the platform against GLBA Safeguards Rule, India DPDPA 2023 cross-border data transfer protocols, GDPR (EMEA assets), Australia Privacy Act, and SOC 2 Type II.

Direct RAG architectures and retrieval patterns using Azure AI Search; plan deprecation/migration from legacy LlamaIndex elements.

Drive cost/performance optimization for Azure OpenAI usage (token budgets, caching, content filtering, model selection/backup strategies).

Partner with DevOps on IaC (Bicep/Terraform), CI/CD, blue-green/canary for Functions and model/prompt releases.

Document and maintain the ownership boundary between platform architecture (this role) and agent runtime implementation (AI/ML Engineer).

Own MLOps practice end to end: dataset curation and golden-set standardization, model/prompt versioning and rollback, offline and online evaluation, drift detection, and CI/CD for model and prompt releases.

Define and lead the roadmap toward Trimont-owned small and large language models, including on-prem/private hosting strategy, fine-tuning and training pipelines on Trimont's document corpus (leases, rent rolls, servicing data), and the build-vs-buy activation triggers (agent volume, per-call cost) that justify the shift.

Design and build the hybrid small/large/frontier model routing strategy — matching extraction tasks to the right model tier by complexity, confidence, and cost — and the multi-model fallback gateway for provider outages or degradation.

Build OnPrem SLM, LLM model training and finetuning.

Coach engineers; perform design reviews and guide root-cause analysis for incidents.

Qualifications:

10+ years in software/ML systems; 3–5+ as AI/ML Design and Build.

Deep Azure experience: Functions, Storage (Blob/Table), Azure AI Foundry + Azure OpenAI, Azure AI Search, App Insights, Key Vault, VNet/private endpoints, APIM.

Production architecture with agentic workflows, RAG, LLM grounding and schema-aligned outputs via Pydantic; strong FastAPI design.

Experience with financial document AI (extraction, classification, or summarization) in a regulated financial context.

Experience operating in a regulated, multi-jurisdiction environment (US financial services + offshore/India delivery) — cross-border data handling, DPA/NDA-gated environments, SOC 2 controls.

Proven governance/observability design (RBAC, audit, replay, evaluation metrics, drift detection).

Hands-on MLOps/LLMOps experience: model/prompt versioning, offline and online evaluation pipelines, golden dataset curation, cost monitoring.

Experience with LLM fine-tuning and model training workflows, and standing up or evaluating on-prem/private small language model (SLM) hosting as an alternative to third-party model providers.

Excellent documentation and stakeholder communication.