Most AI demos stop at Jupyter notebooks.
I build systems that ship.
From YOLOv8 table detection processing 40-page PDFs at production scale, to voice-first AI companions that serve users who can't touch a screen — my work lives in production, not just in .ipynb files.
ElderAI
"I design AI systems around human limitations, not around what's technically convenient."
Parkinson's Detection
"I can take a research problem from architecture design to publication-ready results."
Multimodal Speech Emotion Recognition
"I can architect multi-stream deep learning systems that fuse heterogeneous data modalities."
Signal trace — where I've fired.
GenAI Solution Developer Intern
Built end-to-end vision-parsing pipelines processing 40+ page PDFs with complex table layouts at production scale
Structured data extraction from noisy variable-layout PDFs
Cell-level parsing of detected table regions via Microsoft model
Semantic search pipeline over extracted document content
Multi-format document parsing for downstream LLM consumption
Microsoft Learn Student Ambassador (Beta)
3 workshops delivered to 150+ students, 92% satisfaction rate
Research — IEEE Publication
Parkinson's Detection via Bi-LSTM — 93% accuracy. Paper submitted to IEEE (in review).
Systems in the wild.
ElderAI
Voice-First AI Companion for the Elderly
Elderly people who cannot navigate smartphone interfaces are locked out of the benefits of AI assistants. Existing solutions assume screen literacy.
Multimodal Speech Emotion Recognition
Deep Learning for Virtual Classroom Engagement
Virtual classrooms lack real-time feedback on student engagement. Instructors cannot gauge emotional state from a grid of muted video feeds.
DrowseGuard
Real-Time Driver Drowsiness Detection System
Driver fatigue is a leading cause of highway accidents. Passive monitoring systems are often too slow or inaccurate for high-speed response.
KindKart
Donation-Centric Marketplace for Social Impact
The bridge between excess household items and NGO requirements is broken. Donors lack visibility, and NGOs face massive logistical hurdles in item verification.
RL-Traffic-Lights
Multi-Agent Reinforcement Learning for Urban Flow
Fixed-time traffic signals are inefficient in dynamic urban environments, causing unnecessary congestion and increased CO2 emissions.
RCBS
Production-Grade Indian Banking Simulation
Most banking simulations ignore the complexity of real-world constraints like RBI regulations, transaction idempotency, and multi-agent fraud monitoring.
TrustIQ
AI-Powered Multi-Standard Compliance Assessment
Enterprise compliance across multiple ISO standards (27001, 37001, etc.) is a manual, error-prone process involving thousands of pages of policy documents.
YouTube Doubt Solver
Context-Aware Video Learning Companion
Learners often get stuck on specific concepts in long technical tutorials but lack a way to get immediate, video-contextual clarification.
FoodMind
Production-Grade Nutrition Intelligence
Most health apps offer generic advice. Production systems require high-fidelity data validation, accessible UIs, and hardened security to be viable.
Interview AI-er
AI-Powered Mock Interview Platform
Job seekers lack a low-cost, high-fidelity way to practice interviews with realistic feedback and performance tracking.
Scroll to explore
Open to Summer 2026 internships in AI/ML and GenAI engineering.
garvanand03@gmail.comDesigned & built by Garv Anand — 2025