Worked directly with the founding team on applied AI/ML tasks, contributing to the company's early product development.

What I worked on:

• Assisted in building and testing machine learning components for the product pipeline
• Applied Python-based ML workflows (data preprocessing, model experimentation, evaluation) in a real product environment
• Collaborated in a fast-moving startup setting, translating business requirements into ML-driven solutions

Designed, built, and deployed 4 full production-grade systems end-to-end — from data/model layer to live frontend — covering classical ML, multi-agent LLM orchestration, and AI-driven automation.

What this covers:

• End-to-end ML pipelines: EDA → feature engineering → model training/comparison → explainability (SHAP) → deployment
• Multi-agent AI system design using sequential LLM (Gemini) agents with structured JSON handoffs
• Workflow automation & orchestration with n8n, including self-hosting on Oracle Cloud via Docker
• API integration across Gemini, Serper, and Google Places for real-time data enrichment
• Full-stack delivery: FastAPI/webhook backends paired with React, React Native, and Streamlit frontends, deployed on Vercel, Render, and Streamlit Cloud

AI Automation · Full-Stack Web App

Enter a business type and country — the system searches Google Maps globally with pagination (up to 60 results), scores and filters leads by opportunity, then enriches each one via Serper web search + Gemini AI extraction to pull owner name, email, socials, phone, and website. Results render as interactive React cards with a one-click PDF export.

n8n Google Places API Serper API Gemini API React Vercel jsPDF

🔗 Live Demo · 📂 Source Code

Multi-Agent AI Automation System

Type in a startup idea and watch 4 sequential AI agents go to work: a Skeptic surfaces failure risks, a Researcher finds real competitors via live web search, a Fixer generates counter-strategies, and a Verdict agent scores viability 1–10 with a killer insight. Delivered as a structured 4-section report.

n8n Gemini 1.5 Flash Serper API Streamlit Multi-Agent Orchestration

🔗 Live Demo · 📂 Source Code

End-to-End Machine Learning + Business Intelligence

Full ML pipeline on the IBM Telco dataset (7,043 customers). Four models benchmarked — Logistic Regression, Decision Tree, Random Forest, XGBoost — with XGBoost selected for production (87% ROC-AUC, class-imbalance handling). SHAP explains every prediction. Ships as a 3-page dashboard: Overview, Segment Analysis, and Live Prediction with top-3 SHAP reasons.

Python scikit-learn XGBoost SHAP Streamlit Matplotlib/Seaborn

🔗 Live Demo · 📂 Source Code

ML-Powered Mobile Health Application

Disease detection system trained on 10 years of health data. Multiple models were trained and compared, with Gradient Boosting selected as the best performer. A FastAPI backend serves live predictions to a React Native mobile app.

Python Gradient Boosting FastAPI React Native Render Vercel

🔗 Live Demo · 📂 Source Code