One director. Multiple AI agents. A production-grade platform.
BriefMind is a multi-tenant SaaS platform built by orchestrating AI engineering agents under a single technical vision. This showcase demonstrates what becomes possible when an experienced engineer directs AI at scale — not writing every line, but architecting every decision.
Launch PresentationWhat was built
BriefMind is an agentic automation platform with two major product surfaces: AI Meeting Intelligence (capture, summarize, search across meetings) and Autonomous Data Collection (AI agents that gather information from customers via email, Slack, SMS, and API). The entire platform is multi-tenant, billing-enabled, and production-ready.
Application routes
Infrastructure environments
AI agent capabilities
Architecture highlights
Every architectural decision was deliberate — from the event-driven serverless backbone to the deterministic agent orchestration protocol. This is not a prototype. It is a system designed for production workloads.
AI-Native Architecture
Every layer — from transcript processing to agentic data collection — is designed around LLM capabilities. Not a bolt-on. AI is the architecture.
Serverless Event-Driven
SQS event bus, Lambda functions, and EventBridge orchestration. The system scales to zero and handles bursts without provisioning. Pure cloud-native.
Enterprise-Grade Security
Fernet + KMS encryption for PII, Clerk JWT authentication, multi-tenant data isolation, RBAC, and comprehensive audit logging. Built for regulated industries.
RAG Knowledge Engine
pgvector embeddings with LLM reranking, temporal awareness, and knowledge graph extraction. Semantic search across thousands of meetings with cited answers.
Deterministic Agent Orchestration
AWS Bedrock AgentCore with FLEXY contract protocol. Agents follow deterministic state machines while leveraging LLM flexibility for natural interactions.
Chrome Extension Pipeline
Silent meeting capture — no bot joins the call. Real-time transcription with speaker diarization, streamed to the processing pipeline automatically.
System architecture
Five distinct layers — from the client surface down to the AI and data tier — connected by an event-driven backbone. Each layer is independently deployable and horizontally scalable.
Next.js 15 App
TypeScript + MUI 6Chrome Extension
Meeting CaptureAI SDK React
Streaming ChatFastAPI
Async PythonClerk JWT
SSO + RBACPaddle
Billing + QuotasMeeting Intelligence
Summaries + TasksAgent Orchestration
FLEXY ProtocolRAG Engine
pgvector + RerankingSQS Queues
Inbound / Outbound / RetryLambda Functions
Invoker / Router / SchedulerEventBridge
Cron + RetriesBedrock AgentCore
Agent ExecutionClaude
Summaries + ExtractionMySQL + pgvector
Relational + VectorKMS + S3
Encryption + StorageFull-stack technology map
Frontend: Next.js 15 + MUI 6
TypeScript, App Router, TanStack Query v5, React Hook Form + Zod, TipTap rich text, AI SDK React. 40+ routes with a custom design system.
Backend: FastAPI + SQLAlchemy 2
Async Python with Pydantic validation, Alembic migrations, structured logging, dependency injection. Clean hexagonal architecture with feature modules.
AI Agents: Bedrock AgentCore
Flexible data collector agent, transcript processor, multi-channel orchestration (email, Slack, SMS, Viber). Autonomous operation with human-in-the-loop controls.
Infrastructure: AWS + Terraform
SQS queues, Lambda functions, RDS MySQL, S3, KMS, EventBridge, CloudWatch. Full IaC with dev/staging/prod environments and CI/CD via GitHub Actions.
Documentation pages
CI/CD pipelines
TypeScript + type coverage
How an AI Engineering Director works
The role is not about typing code faster. It is about making the right decisions at the right level of abstraction — then leveraging AI agents to execute across the entire stack simultaneously. The director provides the architectural vision, quality standards, and integration strategy. The AI agents provide the throughput.
Define the Vision
Translate business requirements into technical architecture decisions. Define the product spec, data models, API contracts, and system boundaries. Set the standard for code quality, security, and testing.
Orchestrate AI Agents
Direct multiple AI coding agents in parallel — each handling distinct modules: backend services, frontend components, agent logic, Lambda functions, Terraform infrastructure. Review, refine, and integrate outputs.
Ensure Production Quality
Enforce CI/CD pipelines, type safety, linting, test coverage, and security scanning. Every feature follows the same architecture patterns. No shortcuts, no tech debt accumulation.
Ship & Iterate
Deliver working software continuously. From meeting intelligence to agentic data collection, each capability ships as a complete vertical slice: database, API, UI, tests, and documentation.
AI-directed vs. traditional development
The AI Engineering Director model is not about replacing developers. It is about amplifying a senior engineer's architectural judgment with AI execution capacity.
| Feature | Solo Dev | Traditional Team | AI-Directed |
|---|---|---|---|
| Architecture scope | Limited by one person | Limited by team size | Full-stack, multi-service from day one |
| Consistency | Varies with fatigue | Varies across developers | Enforced patterns across entire codebase |
| Parallelism | Sequential only | Limited by coordination overhead | Multiple agents working simultaneously |
| Documentation | Usually skipped | Inconsistent | Comprehensive — architecture, API, ops guides |
| Time to production | Months for MVP | Weeks for MVP | Days for vertical slices |
| Tech debt | Accumulates fast | Managed with effort | Prevented by consistent patterns |
Capabilities delivered
Meeting Intelligence
Auto-transcription, AI summaries with WYSIWYG editing, action item extraction, and cross-meeting RAG search with cited answers.
Autonomous Agents
AI agents that collect data from customers across email, Slack, SMS, and Viber. FLEXY contract protocol with deterministic state machines.
Multi-Tenant SaaS
Complete org management, team spaces, RBAC, Clerk SSO, and Paddle billing with usage-based quotas and token tracking.
Chrome Extension
Silent meeting capture with speaker diarization — no bot joins the call. Real-time stream processing to the backend pipeline.
Enterprise Security
Fernet + KMS encryption, JWT auth, tenant isolation, audit logging, and infrastructure-as-code with Terraform across multiple environments.
This is what AI-directed engineering looks like
A single engineer with the right vision, the right tools, and AI agents can deliver what traditionally required an entire team. The future of software engineering is not about writing more code — it is about directing intelligence.