Streamline MCP Server Deployment with Python and AWS ECS Express

Latest Project - Fully deployed Sentiment Analysis API built on Azure.   The stack: • Azure Functions (Python 3.11) — serverless compute, no idle cost • Azure AI Language — pretrained sentiment model, no ML training required • Azure API Management — gateway handling routing and policy enforcement • Terraform — entire infrastructure provisioned as code, reproducible and version controlled • Azure Cost Management — budget alerts configured from day one   The API accepts customer review text and returns sentiment (positive/neutral/negative), confidence scores, and extracted key phrases — the kind of data a business can use to track product popularity and customer opinion at scale.   This was a learning project with a real goal: not just getting it to deploy, but being able to explain and defend every architectural decision. I debugged APIM policy tier incompatibilities, route path mismatches, and the distinction between infrastructure provisioning and code deployment — all things that don't show up in tutorials but absolutely show up in production.   Full code and README on GitHub: https://lnkd.in/gvmPDyj7   #Azure #CloudEngineering #Terraform #Serverless #Python #AzureFunctions #DevOps #CloudArchitecture #PortfolioProject

For my bachelor project, AI Code Review, I did not use any cloud API's. Everything ran locally, on a self hosted server, the company owned. Why? - Data sensitivity (company code) - Full control over infrastructure - No external dependencies Setup: - Ollama (local LLM inference) - Python backend (FastAPI) - Async requests + benchmarking - SQLite + validation layer Tradeoffs: - Lower raw model performance - Higher control, security, and predictability In many real-world cases, 👉 control > raw model intelligence #SelfHostedAI #Security #Backend #LLM

🚀 Just Deployed End-to-End ML Project on AWS! Excited to share that I’ve successfully built and deployed a Sentiment Analysis Web Application using Flask and Machine Learning 🚀 🔍 What this project does: It analyzes user reviews and classifies them as Positive or Negative in real-time with a simple web interface. 🛠️ Tech Stack Used: • Python, Flask • Scikit-learn (Logistic Regression, TF-IDF) • NLTK (text preprocessing: stopwords, lemmatization) • Git & GitHub • AWS EC2 (Cloud Deployment) • Gunicorn (Production server) ☁️ Deployment Highlights: • Hosted the app on AWS EC2 • Configured security groups and SSH access • Fixed dependency issues (Windows → Linux environment) • Migrated from Flask dev server to Gunicorn • Managed virtual environments for clean deployment 📊 Project Highlights: • Cleaned and preprocessed real-world review datasets • Built and trained a text classification model • Saved and loaded trained model using .pkl files • Implemented TF-IDF vectorization pipeline • Integrated ML model with Flask backend • Designed simple and user-friendly UI • Handled edge cases like empty input & invalid text • Achieved efficient real-time predictions 🌐 Live Demo: http://54.252.235.253:5000 💡 Key Learnings: • End-to-end ML pipeline (data → model → deployment) • Real-world debugging in cloud environment • Handling version conflicts and dependency errors • Understanding server ports, public vs private IP • Basics of production deployment (Gunicorn) This project gave me hands-on experience with MLOps fundamentals and deploying ML models in real-world scenarios. I’m currently exploring more in AI/ML Engineering & MLOps 🚀 Would love your feedback and suggestions! #MachineLearning #Flask #AWS #MLOps #Python #DataScience #AI #Deployment #GitHub #LearningJourney #CloudComputing #innomatics

🚀 Building a RAG Pipeline on AWS (Step by Step) I’ve started working on a lightweight Retrieval-Augmented Generation (RAG) pipeline, focusing on efficient document processing and embedding. Current workflow: • PDF documents are loaded and split into chunks (600 chars with 120 overlap) • Each chunk is embedded using Amazon Bedrock Titan Embed v2 • Embeddings are stored in ChromaDB (local vector store) • Duplicate detection avoids re-processing existing documents This setup shows how thoughtful chunking and embedding design can create a solid foundation for a RAG system. Next step: I’ll be extending this into a full chatbot interface using FastAPI, backed by AWS Lambda and API Gateway, and deployed via AWS CDK. The final output 🎯 will be the system retrieves relevant document chunks and uses an LLM to generate a meaningful, context-aware response. If you’d like to explore or experiment with the project, feel free to check it out: 👉 https://lnkd.in/d7h8Z235 #RAG #GenerativeAI #AWS #FastAPI #Serverless #MachineLearning #BedRock

Built this over the weekend: a production-ready ML system, not just a model. Most ML projects stop at training. I wanted to go further — and build something that actually runs, evaluates itself, and deploys. Here’s what that looks like: ⚙️ End-to-end pipeline — from raw data → training → versioned models (fully automated) 📊 Automatic model benchmarking — only better models get promoted 🔁 Config-driven system — switch between algorithm pipelines with one change 🌐 Real-time inference — simple UI + FastAPI backend 🚢 CI/CD deployment — Docker + GitHub Actions → AWS ECR → EC2 💡 Key insight: The model is just one component. Designing systems that are reliable, scalable, and self-sustaining — that’s where real ML engineering begins. 🔗 Refer to my GitHub for the complete project Currently focusing on building in this space of production ML systems. #MLOps #MachineLearning #AWS #Python #FastAPI #Docker #SoftwareEngineering

My multi-agent cross cloud article on deploying a Live multi-modal agent to Amazon Lightsail was accepted on the Generative AI channel on Medium!! This paper covers a complete solution that uses the ADK, Python Backend, Websockets, and a Live real-time Gemini Model. The entire system was deployed to an AWS Lightsail endpoint. The Generatative AI channel on Medium is here: https://generativeai.pub/ My article/demo is here: https://lnkd.in/eVBxUGfk #GDE #ADK #A2A #Python #GooogleCloud #AWScommunity #AWS #AWSCommunityBuilders #Lightsail

I just published my first technical blog on community.aws 🚀 Built a fully serverless document Q&A bot using Amazon Bedrock Knowledge Bases — no vector store to manage, no ML expertise needed, answers with citations in under 30 minutes. Inside the post: → Full architecture diagram → Python code (Lambda + Bedrock) → The 3 mistakes I made so you don't have to → Real cost breakdown (~$5/month) This is Part 1 of my series on building production AI on AWS. 🔗 https://lnkd.in/dX3VzzkG #AWS #AmazonBedrock #GenerativeAI #RAG #Serverless #CloudComputing #AWSCommunity

I built and deployed a production RAG system + LangGraph agent to AWS EC2 from scratch. Here is what is inside: 🔹Hybrid BM25 + vector retrieval built from scratch after LangChain deprecated EnsembleRetriever in v1.2 🔹LangGraph research agent with intelligent routing and conversation memory 🔹Automated eval pipeline with JSON result persistence 🔹FastAPI REST API + Streamlit chat UI 🔹Deployed live on AWS EC2 The most interesting challenge: LangChain 1.2 removed EnsembleRetriever. Instead of switching libraries I built custom hybrid retrieval from scratch. That is the kind of problem you only hit in production. Test the live API: https://lnkd.in/eaqQnRqC Full code: https://lnkd.in/eMpnjMdB Built with: Python · LangChain · LangGraph · Llama 3.1 · Groq · ChromaDB · FastAPI · AWS EC2 · Streamlit #RAG #LangGraph #LLM #AIEngineering #Python #AWS #GenerativeAI #OpenToWork

✅ Day 141/365 — Streams, Sorted Lists & Enterprise Features No days off. Here's what went down today: Cloud — AWS Kinesis Data Streams Dived deep into Kinesis Data Streams — one of the core building blocks for real-time data pipelines on AWS. Went beyond just the concept; spun up an actual stream, produced data into it, and consumed it on the other end. Seeing the data flow in real time hits different. LeetCode — Merge Two Sorted Lists Solved it. 0ms runtime — beating 100% of all Python submissions. The approach: collect both lists, sort, rebuild the linked list. Clean and effective. Building — Product is getting serious Today's shipped features: → Role-based access (owner / editor / viewer) → Invite system with real, working links → Code version history (think Git, but lite) → Product is starting to feel genuinely enterprise-level Every day, the gap between where I started and where I'm going gets clearer. 141 days in — no looking back. #365DaysOfCode #AWS #Kinesis #CloudComputing #LeetCode #Python #BuildInPublic #SoftwareEngineering #Day141

I was tired of manually reviewing Terraform PRs… so I built an AI reviewer 🤖 Now every Pull Request: ✔️ Gets a risk score ✔️ Flags security issues ✔️ Suggests fixes inline ✔️ Runs automatically via GitHub Actions Powered by Azure OpenAI + Python. The idea was simple: 👉 Catch risky infrastructure changes BEFORE deployment 👉 Reduce manual effort in code reviews This turned out to be a great hands-on way to explore: DevSecOps AI in CI/CD Terraform best practices Still improving it, but excited about where this can go. Curious — would you trust AI to review your infra code? 🤔 #DevOps #AI #Terraform #Cloud #Azure #GitHub https://lnkd.in/gf6-6JbV