AI-powered SaaS platform reducing audiobook production costs by 95%

Overview

• Founded and scaled AudioFlo.ai from zero to profitability in 90 days, achieving 42% month-to-month growth, 7:1 LTV:CAC ratio, and 35.7% trial-to-paid conversion rate.

Business Impact

• Validated product-market fit through 24 in-depth industry interviews and competitive analysis, defining precise ideal customer profile (ICP) and total addressable market (TAM) opportunity.
• Pioneered a disruptive audiobook business model achieving 95% cost reduction in production. Developed comprehensive go-to-market strategy including product definition and revenue model.

Product & Engineering

• Built accessible audiobook platform with streamlined three-click UX, visual feedback systems, and responsive design, eliminating 85% of production complexity for non-technical users.
• Created breakthrough AI character system that pairs TTS voices with relatable personas and backgrounds, significantly bridging the emotional gap between authors and text-to-speech technology.
• Architected the complete technical infrastructure including frontend, middleware services, database, payment gateway, and notification systems. Designed for cost-efficiency and elastic scaling.
• Developed multi-agent AI system with micro-services architecture and auto-scaling capabilities, automating end-to-end audiobook production pipeline that generated 500+ hours of content.

Growth & Marketing

• Achieved 42% monthly growth through strategic channel selection, targeted content, customer relationship management CRM automation, and data-driven funnel optimization, acquiring 70+ organic customers.

• Led the complete development lifecycle of the flagship text-to-video product, encompassing data pipeline design, model research, validation, serving, and deployment. Delivered on aggressive timeline.
• Developed a patent-pending text-to-video architecture that generates 30-sec videos with unprecedented temporal coherence from just 2 input frames. The system leverages novel data-efficient learning frameworks and achieves breakthrough performance using only 12% of industry-standard training data.
• Created a patent-pending multi-modality video generation system using human pose guidance, seamlessly integrated with synthetic data pipeline to generate 6,500 hours of training data.

• Led a team of 4 ML engineers to develop the 3D human pose estimation module for Metacast Sport Analytics, which drove a 6% stock increase on release day. The ML module, deployed on AWS, served as the fundamental building block of the entire pipeline, processing 50,000+ daily API calls.
• Led 2 ML researchers to build a new 3D human reconstruction method that generates multi-human scenes with complex interactions from only 5 images. The model shipped within 3 months.

• Designed and developed 3 deep learning workflows to tackle lung cancer prediction problems in multi-domain setup with 200 CT images from industrial clients. Results delivered under tight schedules.
• Published 4 peer-reviewed papers in top venues (MICCAI, Gastroenterology, ESMO)
• Architected and delivered 2 critical deep learning micro-services for core product, enabling real-time model training and inference based on user queries.