Over six months as lead technical architect, I designed and built the core of a UAV design optimization platform. The system lets users specify mission requirements — payload weight, flight duration, range, speed — and automatically generates optimized fixed-wing drone designs by selecting from catalogs of real components and optimizing airframe geometry. My contributions spanned the evolutionary search algorithm (genetic algorithm) at the heart of the system, the web API server, the entire AWS cloud infrastructure, and a shared data schema that kept multiple codebases synchronized.
Highlights
- Designed a KD-tree–based component selection mechanism that treats each component as a point in multi-dimensional parameter space, enabling fast nearest-neighbor lookup across large component catalogs.
- Implemented variable-length chromosomes supporting a range of batteries, motors, and sensors with crossover and mutation operators that intelligently handle differing parent component counts.
- Integrated multiple evaluation modules into a penalty-based fitness function that balances cost minimization with mission requirement satisfaction, covering geometric validation, physical simulation, and high-fidelity checks.
- Built the FastAPI backend with Server-Sent Events (SSE) for real-time progress streaming during multi-hour optimization runs, integrated with AWS Batch for compute-intensive workloads and Aurora PostgreSQL for persistence.
- Managed a large Terraform codebase covering core AWS networking, compute, storage, and security services, with S3-backed remote state and GitHub OIDC authentication.
- Established a shared type system (Pydantic + SQLModel) and standardized the dependency management toolchain across all repositories, significantly improving build consistency and resolution speed.
- Coordinated a distributed engineering team across multiple specialized repositories through architecture documentation, OpenAPI specs, code reviews, and regular cross-team syncs.
Technology
- Python 3.11
- FastAPI
- DEAP (Genetic Algorithms)
- scikit-learn (KD-tree)
- SQLModel / SQLAlchemy
- PostgreSQL / Aurora Serverless v2
- AWS Batch · ECS Fargate · ECR · S3 · Secrets Manager
- Terraform
- Docker (multi-stage builds)
- Cloudflare Zero Trust
- GitHub Actions (OIDC)
- UV · AWS CodeArtifact
- Optuna (hyperparameter optimization)
- Alembic (database migrations)