Shebin S Illikkal

🐍

Python

Core Language

📊 DataFlow Automation Suite

A client was spending 3 hours a day moving data between systems manually. Built this to handle the whole thing automatically — pulls from 10+ sources, fixes the messy bits, lands data where it needs to go. Runs every morning without fail.

Processes 2M+ records/day with zero manual intervention

Dynamic schema detection and auto-type-casting engine

Integrated anomaly flagging with email/Slack alerting

CLI dashboard with live progress & error tracking

Python 3.11PandasNumPyCeleryRedis

🤖

Claude AI

AI Expert

☁ SmartAssist Enterprise Platform

Company had 200+ internal docs nobody read. Built a bot using the Claude API that can actually find answers inside those docs. Employees ask it questions in plain English. The HR team didn't believe it would work until they tested it.

Claude API integration with custom system prompts & tool use

Document ingestion — PDF, DOCX, CSV parsed and queried in natural language

Conversation memory with context compression for long sessions

Admin dashboard showing usage, costs, and conversation analytics

Claude APIFastAPIReactLangChainPostgreSQL

📱

FlutterFlow

Mobile App Dev

🛞 ShopEase — Smart E-Commerce App

Client needed a proper shopping app but didn't have a huge budget or months to wait. Used FlutterFlow — it moved fast and the result was solid. Firebase for stock, Stripe for payments. They went from WhatsApp orders to a real app in 6 weeks.

AI product recommendation engine with 35% upsell conversion

Real-time inventory sync via Firebase — zero oversell issues

Stripe & Razorpay dual payment gateway integration

Push notifications, wishlists, reviews & loyalty rewards system

FlutterFlowFirebaseStripeDartPython

📈

TensorFlow

Deep Learning

👁 VisionGuard — AI Security System

Started as a demo I built to test TensorFlow on live CCTV footage. Trained it to spot unauthorized entry and fire. Showed a client, they wanted it installed. It's been running at their warehouse for over a year — caught two real incidents.

Custom YOLO-based detection model trained on 50K+ labeled frames

Real-time 30 FPS inference on edge hardware (Jetson Nano)

Multi-camera dashboard with event timeline & clip playback

SMS/WhatsApp/email alerts within 2 seconds of detection

TensorFlowOpenCVPythonFastAPIMQTT

⚡

React.js

Frontend

📸 AnalytiX — Real-Time Data Dashboard

Client wanted their whole team looking at the same live numbers without emailing spreadsheets around. Built a React dashboard with real-time charts — data updates every few seconds via WebSocket. People actually use it in their morning standups now.

40+ chart types with drag-and-drop widget layout builder

Live WebSocket feeds with sub-second refresh rates

Multi-tenant with role-based access control (RBAC)

PDF/PNG export, scheduled email reports & embed iframes

React 18TypeScriptD3.jsWebSocketTailwindCSS

🔧

Node.js

Backend

🔗 APIForge — Enterprise API Gateway

Backend for a product with multiple services that kept stepping on each other. Built a central gateway with Node — handles auth, routes requests to the right service, and rate limits so nothing gets flooded. It's boring infrastructure that quietly does its job.

JWT + OAuth2 authentication with refresh token rotation

Redis-based rate limiting — 99.98% uptime SLA maintained

Real-time API usage analytics with latency percentile tracking

Auto-generated Swagger docs & Postman collection sync

Node.jsExpressRedisJWTNginx

📊

Power BI

Business Intelligence

💹 BizInsight 360 — BI Platform

Retail client had 50 stores and zero visibility into which ones were actually performing. Built Power BI dashboards pulling from their POS system. Now the owner can see everything in one place from his phone. He checks it every morning apparently.

Live sales dashboards refreshing every 15 minutes via DirectQuery

Custom DAX measures for YoY, MoM, & rolling 90-day comparisons

Regional heatmaps and inventory forecast visuals

Row-level security — each store manager sees only their data

Power BIDAXPower QuerySQLAzure

🗃

PostgreSQL

Database

📌 DataVault CRM — Customer Platform

SaaS startup's database was getting slow and messy as they grew. Rebuilt the schema properly — partitioned the big tables, added the right indexes, rewrote the worst queries. What used to take 4 seconds now takes under 50ms.

Partitioned tables — query speed improved 10x on time-series data

Full-text search with tsvector indexing across all customer notes

Automated backups, point-in-time recovery & replication setup

200+ stored procedures and triggers for business rule enforcement

PostgreSQL 16PostGISpgAdminSQLAlchemyAlembic

🔥

Firebase

Backend-as-a-Service

🔆 ConnectLive — Social Chat Platform

Built a group chat app for a community platform using Firebase. The real-time stuff works really well with Firestore — messages just appear instantly. Got to 10K users before I had to think about scaling, which was a nice problem to have.

Real-time Firestore listeners — messages delivered in <100ms

Firebase Auth with Google, Apple, phone OTP sign-in

Cloud Functions for media processing & spam detection

FCM push notifications with rich media & action buttons

FirebaseFirestoreCloud FunctionsFCMFlutter

☁

AWS / GCP

Cloud & MLOps

🚀 CloudML Deploy — ML Deployment Platform

Client kept asking me to "just retrain the model" every time it drifted. Built a proper pipeline on AWS so the whole thing — retrain, validate, deploy — happens automatically when performance drops below a threshold. Nobody has to ask me anymore.

CI/CD ML pipeline — code push to production inference in 12 min

Auto-scaling endpoints handling 5K RPS burst traffic seamlessly

Model drift detection with automated retraining triggers

Cost optimizer — 60% infra cost reduction via spot instances

AWS SageMakerGCP Vertex AITerraformDockerMLflow

🔗

FastAPI

API Framework

🔴 PredictAPI — ML Inference Service

Needed to serve a few different ML models through one API without slowing everything down. FastAPI was perfect — async, fast, and the auto-docs meant the client's frontend team could work independently without bugging me about endpoints.

Async inference — 3,000+ simultaneous requests without blocking

Multi-model registry with hot-swappable version routing

Auto-generated OpenAPI / Swagger docs with live try-it-out

Request validation, rate limiting & usage metering built-in

FastAPIPythonPydanticUvicornRedis

📄

Pandas

Data Wrangling

📋 DataClean Pro — ETL Pipeline Engine

Fintech client had data coming from 15 different places — different formats, different date styles, duplicates everywhere. Wrote a Pandas pipeline that handles all the cleaning and merging. Runs at 5 AM, and by the time analysts arrive, everything is ready.

Processes 5M+ rows/day — 15 heterogeneous data sources unified

Smart dedup engine with fuzzy matching — 99.3% accuracy

Data quality scorecard with column-level health metrics

Incremental load mode — 80% faster than full-refresh runs

PandasNumPyFuzzyWuzzySQLAlchemyAirflow

🐙

Git / GitHub

Version Control

👥 PyMLUtils — Open Source ML Library

I kept rewriting the same preprocessing and eval helpers across projects, so I packaged them into a proper library and put it on GitHub. More people found it than I expected. Now I maintain it properly — issues, PRs, releases, the whole thing.

GitHub Actions CI/CD — auto-test on every PR, PyPI auto-publish

Semantic versioning with changelogs auto-generated from commits

Branching strategy: Gitflow with protected main + staging branches

600+ stars, 80+ forks, 20+ contributors from 15 countries

GitGitHub ActionsPyPIpytestpre-commit

🛠

Docker

DevOps

📦 ContainerAI — ML Microservices Stack

Needed to deploy a multi-part ML system where each piece could be updated independently without taking everything down. Docker Compose for local dev, Kubernetes in production. First time I deployed a rolling update with zero downtime was a good day.

12-service Docker Compose stack with health checks & restart policies

Kubernetes deployment — rolling updates with zero downtime

GPU-enabled inference containers for deep learning models

Prometheus + Grafana observability stack integrated

DockerKubernetesHelmPrometheusNVIDIA Runtime

🆕

Scikit-Learn

ML Library

💹 ForecastEdge — Sales Prediction Engine

Retail client was ordering stock based on gut feeling. Trained an ensemble model on 3 years of their data. They were skeptical until the first week's predictions came in at 94% accurate. Now they run it before every order cycle.

Stacked ensemble: XGBoost + Random Forest + Ridge — 94% accuracy

Automated feature engineering: lag features, seasonality, promotions

SHAP explainability — which factors drive each store's forecast

Deployed as scheduled job — fresh predictions every Sunday 2 AM

Scikit-LearnXGBoostSHAPOptunaMLflow

🔎

Jupyter

Research & Prototyping

📓 ResearchLab — Interactive EDA Platform

Healthcare analytics project that needed proper, reproducible analysis — the kind where you can see every step. Did all the EDA and statistical testing in Jupyter with full commentary. The client's team could follow the logic without needing to understand Python.

Custom Jupyter extensions for one-click profiling & quality reports

Parameterized notebooks for automated batch EDA runs

Cohort survival analysis & Kaplan-Meier curves on patient data

Git-integrated notebooks with diff-friendly cell outputs

JupyterPandas ProfilingScipyPlotlynbformat

🅾

PyTorch

Deep Learning

🧠 DeepVision — Neural Image Classifier

Trained a vision model on X-ray scans to help flag potential issues for doctors to review. PyTorch gave me the flexibility to try different architectures. The Grad-CAM heatmaps showing what the model was looking at were what convinced the doctors to actually trust it.

Custom ViT + ResNet50 ensemble achieving 97.2% classification accuracy

Grad-CAM heatmaps showing model's region of interest per scan

Training on 120K labeled X-rays with heavy augmentation pipeline

ONNX export for cross-platform edge deployment

PyTorchtorchvisionGrad-CAMONNXAlbumentations

💻

VS Code

Development IDE

🛠 AICode Studio — Smart Dev Environment

Spent too much time setting up new projects from scratch. Built a devcontainer setup with all my usual extensions, settings, and a task that scaffolds a new ML project in one command. Put it on GitHub — a few other people use it now too.

30+ curated extensions: Pylance, Docker, GitHub CoPilot, GitLens

Custom ML project scaffold generator via VS Code tasks

Remote SSH containers for cloud GPU development workflows

Shared workspace settings published as open-source devcontainer

VS CodePylanceDev ContainersGitHub CopilotGitLens

🗼

MongoDB

NoSQL Database

📚 ContentVerse — Headless CMS Platform

Built a headless CMS for an agency running 8 client sites. MongoDB was the right call — content structures vary wildly between clients and I didn't want to do schema migrations every time someone needed a new field. Runs quietly in the background for all of them.

Schema-less dynamic content types — no migrations needed for new fields

Full-text Atlas Search across 500K+ content documents

GraphQL API with DataLoader-based query batching & caching

Change Streams powering real-time preview for editors

MongoDB AtlasMongooseGraphQLNode.jsRedis

🚀

Postman

API Testing

✅ APIGuard — Automated Test Suite

Fintech project where breaking an API endpoint was a real problem. Built out a full Postman collection covering 400+ endpoints with proper test assertions. Plugged it into GitHub Actions so every deployment gets tested before it touches production.

400+ test cases across auth, CRUD, edge cases & error flows

Newman CLI integration — tests run in GitHub Actions on every PR

Environment-aware collections: dev / staging / production configs

Automated contract testing — catches breaking API changes instantly

PostmanNewmanGitHub ActionsJavaScriptChai

A bit about me

Claude AI & Cloud — Solid Knowledge

Technical Skills

Machine Learning & AI

Python & Libraries

App Development

Frontend Development

Backend Development

Cloud & AI Tools

What I do

Machine Learning & AI

Data Analytics & BI

Mobile App Development

Full-Stack Web Development

Claude AI & Cloud Integration

Data Science Consulting

Some of my work

AI Analytics Platform

ML Sales Forecasting

E-Commerce Mobile App

Customer Churn Prediction

Claude AI Support Bot

Automated Data Pipeline

Tools I actually use

Get in touch

Drop me a message

Message sent!