About Caffeinated Code
.jpg/:/cr=t:1.96%25,l:0%25,w:100%25,h:96.08%25/rs=w:1024,h:1365,cg:true)
My Development Journey
I still remember the first time I stared at a blank screen, fingers on the keyboard, wondering if I could make something appear from nothing but keystrokes. That was over twenty years ago, and the thrill of seeing an idea come alive in code never left me.
Over the years, I kept chasing that spark—teaching myself, breaking things, fixing them, and slowly learning how to think like a builder. When I went through Columbia College, I wasn’t just checking boxes; those courses in Agile Software Development, Web Programming, Data, and Databases gave me a framework to sharpen my instincts. They taught me to break problems down, to work in iterations, to let data guide the path forward.
Today, what keeps me hooked is the same thing that drew me in at the start: solving real problems with working software. I’ve built tools across web, mobile, data, and machine learning, and while the technologies have changed, the joy hasn’t. I love the craft—the small choices that make software dependable, the feedback loops that turn rough edges into polished systems, and that moment when something I built makes another person’s job easier or faster.
The toolbox has grown—Python, C#, ASP.NET MVC, Git, Firebase, SQL, AWS, GCP, React, Tailwind, Django, Docker, Kubernetes, ML models—but at its core, it’s still about the same thing: taking an idea and turning it into something real.
My coding journey...
All-Star Connect
Purpose: Team chat, announcements, calendars, attendance, media sharing.
- Stack:
- Mobile: Expo React Native
- Web: Next.js coach console
- Backend/DB: Firebase (Auth, Firestore, Storage)
- Notifications: FCM (topic + token targeting)
- Scheduling/Jobs: Cloud Functions + Tasks
- Media: Proxy transcodes via FFmpeg
- Note: Shares auth/roster with the gym platform for single sign-on.
Coach Vision
Purpose: AI-Enhanced Practice Review. Cut 60–90 min practices into 8–12 min highlights; remove dead time.
- Stack:
- Ingest/Store: Firebase Storage, metadata in Firestore
- Processing: FFmpeg, Python (OpenCV for motion; MediaPipe/MoveNet for pose; audio energy via librosa/webrtc)
- Orchestration: Cloud Functions, Cloud Tasks (queue), optional GPU on GCP
- Frontend: Next.js (coach web), Expo (mobile)
- Output: MP4 highlight + chapters JSON; timeline heatmap UI
Tumbling Video Analysis
Purpose: Athlete uploads clip of tumbling. → get deductions, scores, and drills.
Stack:
- Ingest: YouTube URL → proxy via FFmpeg
- Vision: MediaPipe/MoveNet for 2D pose; optional BlazePose; smoothing with One-Euro filter
- ML/Scoring: Python; rule-based kinematics + optional PyTorch/TF models (phase detection, angle thresholds)
- Backend: FastAPI (Python) behind Cloud Run or Cloud Functions (HTTP)
- DB: Firestore (attempts, scores, drills)
- Frontend: Next.js/Expo; overlay skeleton + angle charts
- Note: Starting rule-based; train ML models later from tagged library. Currently utilizes OpenAI.
Trucking Dispatch Board → Multi-Tenant AWS
Purpose: Drivers, loads, routes, statuses; customer/shipper portals.
Stack (C# path):
- Web: ASP.NET Core MVC
- Auth: AWS Cognito (org/tenant claims)
- DB: PostgreSQL (RDS) with Entity Framework Core; multi-tenant via schema or discriminator
- Caching/Queue: Redis (ElastiCache), SQS
- Storage: S3 (PODs, BOL PDFs)
- Deploy: ECS Fargate or Elastic Beanstalk; CI/CD GitHub Actions
- Frontend UI: React SPA served by .NET
BlendIQ
Purpose: Formulate chemical blends, model costs, produce tech sheets.
Stack:
- Backend: FastAPI (Python) for calc engine (unit conversions, density/viscosity models)
- Frontend: React/Next.js
- DB: Postgres (blend recipes, components, SDS links)
- Exports: Pandas → CSV/XLSX; Puppeteer → PDF
- Note: Add a versioned formula history table for auditability.
Recidivism Risk Scoring Tool
Purpose: Predict likelihood of re-offense → for research/policy (not production law enforcement).
- Data: CSV/SQL datasets (e.g., prior offenses, age, socioeconomic markers).
- Modeling: Python (scikit-learn, TensorFlow/PyTorch) logistic regression + ensemble models.
- Bias Mitigation: AIF360 fairness toolkit.
- API: FastAPI (score endpoint).
- Dashboard: Streamlit or Next.js with visualization (probability scores, factors).
- Opinion: Great “social good” project + portfolio piece; make fairness/explainability central.
Online Quote System
Purpose: Online booking + instant pricing by lot size for lawn/painting jobs.
- Frontend: React/Next.js widget with TypeScript + TailwindCSS, embedded into WordPress.
- Backend: Firebase Cloud Functions (Node.js/TypeScript) to handle pricing logic, API calls, and submissions.
- Database: Firestore NoSQL for storing service options, lot-size pricing tables, and customer requests.
- APIs: Google Maps (Geocoding, Places, Distance Matrix) + Estated/ATTOM property parcel APIs for lot size detection.
- Payments (Optional): Stripe Checkout integration (JavaScript SDK on client, Node.js backend).
- Embedding: Minimal PHP inside WordPress to host the booking widget; main logic runs client-side in React.
Resume Analyzer
Purpose: Upload a resume → parse, analyze, and return feedback on strengths, weaknesses, and keyword alignment for ATS (Applicant Tracking Systems).
- Frontend: React/Next.js (TypeScript + TailwindCSS) with a simple drag-and-drop resume upload and results dashboard.
- Backend: Python (FastAPI) for parsing, text cleaning, and analysis pipeline.
- Database: PostgreSQL (structured storage for resumes, parsed metadata, and scoring results).
- APIs & Libraries:
- NLP: spaCy + NLTK for parsing and keyword extraction.
- ML: scikit-learn + Hugging Face Transformers for classification and scoring.
- PDF/DOCX Parsing: PyPDF2 + python-docx.
- Embedding/Deployment: Dockerized app running on AWS (ECS + RDS). CI/CD with GitHub Actions.