30 things I've shipped

Chrome extension that catches cognitive biases in real-time BEFORE you make bad decisions. Integrates Claude API for AI-powered analysis. Detects Sunk Cost Fallacy, Confirmation Bias, Planning Fallacy, Loss Aversion, and more with reframe suggestions.

Fine-tuned Qwen3-4B using QLoRA with 4-bit NF4 quantization for resume enhancement. Achieved 9.5/10 quality scores with 18-22GB GPU memory. Processed 1800+ resumes and built REST API with FastAPI.

Kubernetes-based ML platform handling 500 RPS with circuit breakers, HPA scaling, and Prometheus monitoring. 70% cost reduction with spot instances. Real-time toxicity detection with 6 classification labels.

Production MLOps pipeline with 8 microservices on AWS ECS Fargate. SageMaker training, DynamoDB storage, automated retraining with SNS triggers. Complete ML feedback loop with auto-deployment.

Enterprise serverless architecture with 94 Lambda functions. Stripe Connect payments, Twilio SMS, DocuSign e-signatures, QuickBooks accounting, and EagleView imagery integrations.

Multi-tenant RBAC backend in Go with Gin framework. JWT auth, DynamoDB GSI/LSI, real-time sync for field service management. Complete user lifecycle with role management.

Autonomous object cluster detection using ROS services. Multi-strategy arm manipulation with YOLO + CLIP segmentation. Point cloud processing and farthest point sampling algorithms.

Traffic prediction using Graph Neural Networks on PeMS-BAY dataset. Spatio-temporal modeling with PyTorch Geometric. End-to-end pipeline from SUMO simulation to anomaly detection.

AI agent for commercial real estate research using RAG with Qdrant vector DB. Multi-step pipeline with embeddings and reranking. Discovers ~20 relevant papers with inline citations.

Serverless backend using FastAPI, AWS Lambda, and DynamoDB. Multi-LLM pipelines (GPT-4, Gemini, DeepSeek) with custom prompts. Sub-2s PDF generation and 80% reduction in manual effort.

Multimodal pipeline combining Reddit sentiment (VADER + virality scoring) and financial indicators. 7-day rolling LSTM model in SageMaker. Predicted NVDA next-day returns.

Newsletter automation using n8n (no-code) and CrewAI + LangChain (code-first). Compared performance and flexibility. 70% reduction in manual workload with multi-agent coordination.

Pneumonia detection using VGG19 with custom classifier layers. Transfer learning and image augmentation for medical imaging. Flask web app with real-time inference.

25+ HATEOAS-compliant endpoints with OpenAPI docs. Full job lifecycle management with PDF resume uploads, duplicate prevention, and multi-parameter filtering. Deployed to AWS ECS.

Field service management with 104+ TypeScript components. Real-time VOIP calling (Telnyx WebRTC), PDF generation, job assignment, time tracking, and Stripe payments.

Compared 6 HPO methods for CNN. ES(1+1) achieved 79.80% vs 72.47% baseline (+10.1% improvement). Implemented CMA-ES, Hyperband, DEAP optimization algorithms.

6 microservices with Jenkins Multibranch Pipeline. Branch-per-service CI strategy with independent Docker builds. Kubernetes manifests with LoadBalancer for API gateway.

VSCode-themed portfolio with editor aesthetic. GitHub contribution calendar, theme switching (GitHub Dark, Dracula, Ayu, Nord), and dev.to article integration.

Newsletter automation with AI agents. SerpAPI for trending topics, Tavily for research, Ollama local LLM (Llama 3.2), and Qdrant vector DB for semantic search.

Stock prediction using hybrid sentiment analysis. Combined financial data with NLP-based scoring. Text preprocessing, NER, and ML models for market prediction.

Web scraper for job descriptions. Multi-format resume parsing (PDF, DOCX, TXT). Generated 1,304+ training examples with ATS score calculation and parallel processing.

3-tier architecture: Nginx frontend, Flask API, PostgreSQL. Kubernetes with 3 deployments, health checks, and PVC. 10+ REST endpoints for car inventory management.

FastAPI proxy server for aviation API. 11 endpoints (flights, routes, airports, airlines). OpenAPI 3.0 specification with Swagger UI documentation.

Novel dual step-size system with (μ+λ) population strategy. Covariance-based adaptation learning dimension correlations. Outperformed baseline ES algorithms in competition.

Time series forecasting using LSTM and GRU architectures. UCI Air Quality Dataset analysis. Comprehensive comparison of recurrent architectures for prediction.

6 AWS ML projects: MNIST CNN, Keras HPO, Transformers, TF-IDF, X-Ray imaging, Kinesis streaming. Distributed multi-GPU training with CloudFormation IaC.

Cross-platform Flutter app with ESP32/NodeMCU control. MQTT-based Adafruit Cloud sync, RGB light control, and Google Assistant + Alexa voice integration.

Real-time COVID-19 dashboard using disease.sh API. Country selection with persistence, educational content sections, and responsive UI with image caching.

Inventory Management REST API built with Java 21 and Spring Boot 3.5. Clean Controller→Service→Repository architecture, paginated search with auto stock status detection (IN_STOCK/LOW_STOCK/OUT_OF_STOCK), Jakarta validation, and JUnit 5 + Mockito unit tests. Containerized with Docker Compose.

10.81M-parameter GPT decoder-only transformer built from scratch in PyTorch. Custom BPE tokenizer via SentencePiece achieving 3.11× compression. Trained on dual RTX 3090 GPUs reaching val loss 1.18 at 25K steps. Deployed as Gradio app on HuggingFace Spaces.