I am a BCA graduate and lifelong learner currently specializing in Machine Learning and Deep Learning.
My journey started in Full-Stack Web Development, and I am now focused on building intelligent systems, generative models, and NLP solutions.

• 🔭 Currently: Enrolled in an intensive Machine Learning program at Brototype
• 🧠 Focusing on: Generative AI, LLMs, Computer Vision, and MLOps
• ✍️ Technical Writing: I simplify complex AI concepts like Transformers and GANs on my Medium blog
• 📍 Location: Kozhikode, Kerala, India

LingosAI is a full-stack, AI-driven English tutoring platform built for non-native speakers who want career-ready communication skills. It runs a structured multi-week curriculum through a real-time, chat-driven lesson interface, where every session follows a teach → task → evaluate → feedback lifecycle. An agent-based AI layer generates personalized writing, speaking, reading, and listening activities based on each learner's diagnosed weak points, while a RAG-powered feedback memory system lets the AI tutor recall a learner's recurring mistakes across sessions. Scoring is handled by a separate deterministic engine (no LLM involved), keeping progress tracking reproducible and explainable across seven measurable English sub-skills. The platform is live in production with a subscription model, admin dashboard, and a fully automated CI/CD pipeline from pull request to deployment.

🔗 Live Site: lingosai.com
💻 GitHub: github.com/orbin123/lingos-ai

Key highlights:

• Backend: FastAPI, SQLAlchemy 2.0, Alembic, JWT + Google OAuth, role-based access control
• Frontend: Next.js 16 (App Router), React 19, TypeScript, Tailwind CSS v4, WebSockets-driven chat UI
• AI Layer: LangChain + LangGraph multi-agent system (teacher, evaluator, feedback, task-generator, planner), RAG feedback memory via Pinecone, provider-agnostic AI interfaces (LLM/TTS/STT)
• Infra: AWS ECS Fargate + ECR, RDS (PostgreSQL 16), Redis, S3 + CDN, Terraform (IaC), Vercel (frontend)
• CI/CD: GitHub Actions — lint, type-check, unit + integration tests, migration replay, coverage gate, Docker + Trivy CVE scanning, automated rolling deploys with smoke-test rollback

• MNIST CNN Deployment with AWS SageMaker
  End-to-end PyTorch pipeline — custom CNN trained with augmentation and LR scheduling, packaged and deployed as a live SageMaker inference endpoint (~99% test accuracy).

• Boston House Price Prediction API — CI/CD & Monitoring
  Production-style ML API with Flask + Random Forest, containerized via Docker Compose alongside Prometheus and Grafana, with a full GitHub Actions pipeline for testing, evaluation, and automated Docker builds.

• FinBird: Real-Time NLP for Financial News Intelligence
  Hybrid Django + FastAPI system applying NER, LDA topic modeling, and Transformer-based summarization, with sentiment analysis benchmarked across VADER, BERT, and LSTM.

• Telco Customer Churn — Spark ETL & ML Pipeline
  End-to-end PySpark ETL pipeline extracting, transforming, and validating data into optimized Parquet datasets, with feature engineering across 20+ attributes feeding Logistic Regression and Random Forest models.

• CycleGAN: Sketch-to-Photo Translation
  Unpaired image-to-image translation using ResNet-based generators and PatchGAN discriminators, evaluated with FID scores.

• 📖 Attention Is All You Need (In Simple Terms)
• 📖 Hello to Generative Models: VAE & GAN
• 📖 CycleGAN for Converting Face Sketches to Real Images
• 📖 Cloud ML Deployment: SageMaker, Vertex AI & Azure ML