Portfolio

End-to-end point-of-sale platform with customer ordering, staff roles, and inventory management. Built a normalized PostgreSQL schema with triggers/constraints, transactional order processing, and audit logs. Implemented a responsive JavaFX UI with role-based access, receipt printing, and admin dashboards.

Backend uses a DAO layer over JDBC, batch inserts, and parameterized queries for safety and speed. Deployed on AWS with environment-based configs and GitHub Actions for CI build checks.

Full-stack prototype that turns a calibration run into a defensible compute decision. Users upload an ML repo, run a sandboxed job in Docker, and AURORA captures runtime, logs, and profiling signals (CPU/RAM/IO) to estimate scaling and recommend fastest / cheapest / balanced options with clear reasoning.

Produces reproducible run bundles (Dockerfile + scripts + README) to bridge local testing → deployment. Optional GenAI-assisted calibration suggestions and chat for quick tuning and diagnosis.

Production-style event-driven analytics pipeline: ingests high-volume service events (latency, status, timestamps), streams them through Kafka-compatible Redpanda, computes rolling-window KPIs, persists results to PostgreSQL, and visualizes operational health in a near real-time React dashboard.

Built as a clean multi-service repo with containerized local infrastructure and a streaming worker that performs aggregation and durable KPI writes.

Internal tool that turns raw event-level data into business KPIs: revenue (sales − refunds, stored safely in cents), activity, daily trends, and event-type breakdowns. Includes strict role-based access control: admins see platform-wide metrics; users see only their own.

Node/Express backend with JWT auth and backend-enforced RBAC, backed by PostgreSQL with indexed time-series queries for fast KPI aggregation. React UI keeps the dashboard clean and readable.

Implemented a Windows traceroute tool that sends TTL-limited ICMP probes, resolves hop names asynchronously, and prints ordered path results with RTT timing and ICMP error handling. Includes raw-socket mode plus Windows ICMP Helper API fallback for restricted environments.

Batch mode traces many targets with worker threads and a topology-aware smart mode that reuses stable prefixes. This originated from CSCE 612 coursework; the public repo includes an academic integrity notice.

Built a custom reliable sender over UDP with connection setup, packet sequencing, ACK processing, sender-window flow control, retransmission timers, RTT/RTO updates, duplicate-ACK fast retransmit, and graceful FIN close.

The sender coordinates producer and background ACK threads to keep the transfer window full while tracking throughput and checksums. This originated from CSCE 612 coursework; the public repo includes an academic integrity notice.

Implemented a command-line DNS client that builds raw DNS queries, sends them to a chosen resolver, and decodes answers, authority records, and additional records from the response packet.

Supports A and PTR lookups, compressed-name parsing with bounds and loop checks, retries, timeouts, and cross-platform sockets. This originated from CSCE 612 coursework; the public repo includes an academic integrity notice.

Multi-objective routing that balances time, fuel, and operational risk. The engine fuses live weather and port conditions with a news-derived risk score (Transformer sentiment + rules) to adjust edge weights in real time. Constraints like draft limits, traffic separation schemes, exclusion zones, and canal fees are modeled explicitly. The policy stack combines a Transformer forecaster with PPO-based routing under uncertainty.

A Streamlit UI shows fastest / cheapest / safest alternatives with deltas in ETA, fuel burn, and risk. Each route includes short LLM rationales. The pipeline includes cached data fetchers, reproducible configs, and scenario "playbacks" for incident post-mortems and what-if planning. Runtime services are monitored via Prometheus and Grafana.

Backend-first semantic search MVP: ingest images into an embedding index, embed text + images into the same CLIP vector space, and return ranked results via cosine similarity. Exposes clean endpoints for health checks, ingestion, search, and Prometheus-compatible metrics.

Designed to be explainable and interview-friendly: simple in-memory vector store (easy to swap for FAISS/pgvector), containerized dev with Docker Compose, and basic observability dashboards.

Built a simulation-based risk engine using log returns and distribution diagnostics to justify modeling assumptions. Runs Monte Carlo paths to estimate VaR, CVaR, drawdowns, and scenario stress tests.

Modular data loaders + plotting utilities, with a clean "main" entry-point for quick experiments and parameter sweeps.

Controlled benchmark comparing a compact CNN to a Vision Transformer under limited data regimes. Tracked scaling behavior across dataset fractions, seeds, and augmentations; exported metrics and checkpoints for reproducible comparisons.

Includes training curves, calibration checks, and error analysis tooling to understand when transformers win (and when they don't) without hand-wavy claims.

Classic VGG19 neural style transfer with a streamlined UI: upload content & style, then tune content/style weights, TV loss, and iteration budget. Uses a coarse→fine schedule and multi-scale Gram features for crisp textures without over-sharpening.

Includes deterministic seeds, progress snapshots, and side-by-side previews. Runs on CPU or GPU; large images are tiled and reassembled to keep memory bounded and results reproducible across devices.

Based on "A Neural Algorithm of Artistic Style" →

Transparent fake-review classifier that separates CG (computer-generated) vs OR (organic) reviews. Built to be lightweight and explainable—no black-box embeddings required.

Uses hand-engineered features (length, suspicious phrase count, ALL-CAPS word count) with standardization, then trains an Adaline model via batch gradient descent and MSE loss in NumPy.