Rohith EP
Back to Selected Work
AI Application · Internship·Mexilet Technologies·2026

AI Property Video Generator

Automated AI platform that converts property URLs, brochure PDFs, voice notes, text descriptions, and images into narrated, ready-to-publish real-estate marketing videos.

Role

AI & Backend Engineer Intern

Organization / Client

Mexilet Technologies

Year

2026

Live URL

makepropertyvideo.in
AI Property Video Generator

The Problem

Real-estate teams and brokers spent hours manually extracting listing details from brochures or websites, writing scripts, recording voiceovers, sequencing room photos, and editing walkthrough videos.

The Engineering Approach

Engineered a three-phase automated production pipeline on Frappe and MariaDB—combining OpenAI Whisper for voice-note transcription, Amazon Bedrock for AI narration scripts, ElevenLabs for lifelike Text-to-Speech, CLIP & FFmpeg for visual sequencing and video assembly, and Razorpay for seamless regional payments.

System Architecture & Workflow

Phase 01

Multi-Modal Data & Audio Extraction

Processes raw user text descriptions into structured property data and transcribes voice notes using the OpenAI Whisper speech recognition model (chosen over Google STT and Azure Speech for superior multilingual accuracy). Paired with Selenium for dynamic URL extraction and Python PDF parsing for property brochures.

Phase 02

AI Narration & Lifelike Speech Synthesis

Generates engaging, scene-by-scene property narration scripts via Amazon Bedrock and converts scripts into realistic voiceovers using ElevenLabs Text-to-Speech. Property photos are classified and ordered logically (Exterior → Entrance → Living Room → Bedroom → Kitchen) using the CLIP vision-language model.

Phase 03

Frappe Backend, FFmpeg Rendering & Razorpay

Built on a modular Frappe framework backend with MariaDB storage. Integrates Razorpay (UPI, cards, and net banking) for pay-per-video checkout and uses FFmpeg to combine ordered visuals, ElevenLabs narration audio, and property metadata into a finished marketing video.

Key Contributions & Deliverables

  • Structured text extraction & OpenAI Whisper multilingual audio transcription pipeline
  • Automated property narration script generation using Amazon Bedrock LLMs
  • Natural voiceover synthesis integrated via ElevenLabs Text-to-Speech (TTS)
  • Razorpay payment gateway integration supporting UPI, cards, and net banking
  • End-to-end workflow orchestration on Frappe Server with MariaDB, CLIP room ordering, and FFmpeg rendering

Technology Stack

PythonFrappeMariaDBAmazon BedrockElevenLabs TTSOpenAI WhisperCLIPFFmpegRazorpay