Problem: Businesses need 24/7 receptionist capabilities without the cost of human operators. Manual booking over phone is error-prone and unavailable after hours.

Solution: A real-time voice AI that manages bookings through natural conversations using a speech-to-LLM-to-speech pipeline. Maintains conversation context across turns, confirms reservations, and sends automated email confirmations.

Role: Solo developer — designed architecture, implemented STT/TTS pipeline, integrated LLM for intent extraction and slot filling, built email automation.

Architecture: Audio Input → STT (Whisper) → LLM (intent + slot filling) → TTS → Email + Booking Storage

Tech: Python, FastAPI, OpenAI, Whisper, TTS, Prompt Engineering, SMTP

Problem: Job seekers struggle to identify the right career paths based on their existing skills, often missing opportunities that align with their strengths.

Solution: A semantic matching system using TF-IDF and cosine similarity to align resumes with job descriptions. Built a backend service mimicking RAG patterns by retrieving relevant skill documents before inference, connected to a Streamlit frontend.

Role: Solo developer — implemented resume parsing, feature engineering pipeline, job similarity scoring algorithm, and recommendation engine.

Architecture: Resume Parsing → Feature Engineering → Job Similarity Scoring → Role & Project Recommendations

Tech: Python, pandas, NumPy, scipy, scikit-learn, EDA, Feature Engineering

Problem: Estimating real estate prices manually is subjective and inconsistent. Buyers and sellers need data-driven valuations to make informed decisions.

Solution: End-to-end ML pipeline for predicting house prices using regression models and feature engineering, with interpretability to explain which factors drive property values.

Role: Solo developer — built complete pipeline from data ingestion through feature engineering, model training (Ridge/Lasso regression), evaluation, and Streamlit frontend.

Architecture: Data Ingestion → Feature Engineering → Model Training (Ridge/Lasso) → Evaluation → Prediction Output

Tech: Python, pandas, NumPy, scikit-learn, Ridge & Lasso Regression, Streamlit