Forward Deployed Engineering

Install Docker Desktop with its required Windows virtualization support and verify the engine by running a test container.

Define Pydantic models for a user request, a cited evidence item, and a typed decision response.

Call the Tavily search API and normalize its results into a consistent evidence structure.

Implement an Amazon Bedrock invocation function with Boto3 and validate request and response handling with a stub or mock when sandbox credentials are unavailable.

Implement a provider-neutral typed decision adapter that validates decision, probability, and confidence fields and routes low-confidence results for human review, binding it to Jev only through verified official documentation.

Compose Tavily search, Bedrock synthesis, and Jev decision steps into a local Python workflow.

Expose the workflow through a FastAPI POST endpoint with defined responses for invalid input, missing configuration, timeouts, and provider failures.

Write automated tests for successful, low-confidence, and provider-failure paths by replacing external services with mocks.

Use Docker Compose to start local PostgreSQL and Weaviate services and verify their health before the application connects.

Design and apply a PostgreSQL schema for requests, evidence references, model outputs, decisions, and processing status.

Insert and query workflow records from Python using parameterized SQL and explicit transactions.

Create a Weaviate collection, insert evidence with test vectors and metadata, and run a similarity search with a metadata filter.

Combine PostgreSQL workflow records with evidence retrieved from Weaviate for a single API request.

Containerize the FastAPI application and add it to Docker Compose with environment-based configuration and a health check.

Choose between Lambda and ECS Fargate for a workload using duration, scaling, packaging, networking, and cost constraints.

Draw the selected AWS architecture across API Gateway, compute, Step Functions, Bedrock, RDS for PostgreSQL, Weaviate, Tavily, and the decision adapter, clearly marking trust and network boundaries.

Configure the AWS CLI for temporary sandbox credentials and verify the active account, role, and region.

Grant a compute workload least-privilege access to Bedrock and other required AWS resources with an IAM policy.

Store external API credentials in AWS Secrets Manager and retrieve them at runtime without exposing them in logs.

Use a repeatable runbook to deploy the selected compute path through API Gateway, using Lambda or ECR and ECS Fargate according to the decision made on Day 3.

Implement a Step Functions state machine that sequences tasks and uses retry, catch, and timeout behavior.

Design a VPC topology for isolated compute and RDS resources using subnets, routes, security groups, and controlled outbound access, and identify where AWS Network Firewall would enforce egress policy.

Add request correlation identifiers to CloudWatch logs and diagnose a seeded failure by separating application, IAM, networking, configuration, and service-quota causes.

Deliver a five-minute demonstration that exercises a successful request and a controlled failure, then justify the design in terms of security, reliability, scaling, latency, and cost.