Principal Architect (AI)

Brightside

About The Position

We are looking for a Principal AI Architect to own the design and build of Brightside’s AI agent platform. This is a senior, hands on role. You will work directly alongside Jacky Chiu and the engineering team to architect and ship production systems, not design from a distance. The core system you will build is a realtime agent layer that listens to live conversations between Financial Assistants and clients, understands what is happening turn by turn, surfaces intelligent coaching to the FA in under 2 seconds, and writes structured data back to the CRM. As the platform matures, this same agent layer will conduct client conversations autonomously. This role is execution oriented and involves designing and building systems, not governance, advisory, research, or people management.

Requirements

10+ years in software engineering or systems architecture, with significant hands-on experience in the last 3 years.
Production experience building real-time AI systems, specifically systems that operate on a per-turn, sub-second latency budget.
Hands-on experience with Amazon Bedrock, AWS Lambda, and Kinesis (have hit their limits and worked around them).
Multi-agent system design experience (orchestration patterns, agent handoff, context sharing, structured output validation).
Strong data architecture foundation (relational modeling, event-driven systems, CRM data patterns).
Experience in a regulated environment (fintech, financial services, or healthcare) with an understanding of privacy and compliance requirements.
Comfortable operating in a small, fast-moving team where you design and build.

Nice To Haves

Experience building voice or conversation intelligence systems (contact center AI, real-time transcription pipelines, live agent coaching tools).
Hands-on experience with Amazon Connect and Contact Lens.
Background in conversational AI product companies (Cresta, Observe.AI, Cogito, Replicant, or similar).
Experience with LLM prompt engineering at a systems level (eval harnesses, versioned prompt governance, regression testing).
Prior experience as a lead architect at a Series B-D company where you were one of a small number of senior technical voices.

Responsibilities

Design and own the multi-agent orchestration layer, including session supervisor, domain clusters, escalation arbiter, and conversation manager.
Define agent routing logic, context sharing, tool use, and latency budgets across 18+ agents.
Architect the real-time pipeline: Amazon Connect → Contact Lens → Kinesis → Lambda → Bedrock → WebSocket push to FA panel.
Own the dual-mode agent architecture: human-assisted mode (agent coaches the FA) and autonomous mode (agent conducts the client conversation directly).
Define model selection and routing across low-latency and reasoning-intensive agent tiers.
Govern the prompt architecture: structured agent prompt standards, runtime domain knowledge injection, and conditional output formatting across delivery surfaces.
Build and own the eval harness for schema validation, behavioral regression testing, and latency profiling.
Design the domain knowledge base structure across the full financial use case library (8 domains, 49 use case types).
Own the data model underlying the AI platform: client financial profiles, cases, goals, options, outcomes, and financial impact measurement.
Architect the CRM sync layer, including real-time field writes after FA confirmation and human correction logging as a training signal.
Define canonical data models across clients, employers, financial products, and outcomes.
Ensure data integrity across Aurora PostgreSQL (new CRM) and Aurora MySQL (existing platform).
Own the AWS Bedrock-native inference infrastructure, including async invocations, Knowledge Bases, and RAG pipelines.
Design the session state layer using ElastiCache Redis for hot state and DynamoDB for durable session records.
Define AI governance standards: model evaluation, monitoring, explainability, and compliance in a regulated fintech environment.
Evaluate and guide decisions on model providers, orchestration frameworks, and platform tooling.