Hey, I’m Prayag.
Building AI systems that work in the real world — from retinal diagnostics to Alzheimer’s drug discovery.
Years of Learning
& Building
Research
Projects
Hackathon
Wins
I got hooked on AI when a simple image classifier I built in second year actually worked on real hospital data. That moment — seeing code make a difference outside a textbook — changed everything for me.
Since then, I have been chasing that same feeling: building systems that solve problems people actually face. From detecting particles inside Bisleri bottles on a factory floor to predicting drug activity for Alzheimer's research, I care most about work that ships and works in the real world.
When I am not training models or debugging inference pipelines, you will find me at hackathons — 7 wins and counting — or mentoring juniors on how to go from "it works on my laptop" to "it works in production."
Years of Learning
& Building AI
Research
Projects
CGPA — B.E. AI & Data Science
Strategic Depth,
Real Execution
Industry Collaboration with Bisleri
Designed a two-stage vision pipeline with NIR illumination and a controlled dark-tunnel to detect >500 µm particles inside transparent bottles. Achieved ~98% accuracy with 1–2% false alarms.
Affordable Diagnostic Tool for Rural India
Led development of a MobileNetV3 model trained on 85,000 retinal images, achieving 93% accuracy. Designed for deployment on low-cost hardware to serve rural practitioners — model is live and accessible beyond the research group.
Alzheimer's Drug Discovery
Built a regression model predicting pIC50 values for Alzheimer's target inhibition. Extracted molecular fingerprints with RDKit, optimised LightGBM with Optuna (100 trials), achieving CV R²=0.87 — up from RF baseline of 0.77.
AI Posture Correction for Mobile · Patent Filed
Designed and deployed a full posture detection pipeline using YOLO on TensorFlow Lite. Improved pose detection accuracy from 68% to 94.45% on mobile hardware. Resolved class imbalance across 12 posture categories.
Automated Structural Inspection
Developed a MobileNetV3 model for automated crack detection and severity grading on construction walls. Replaced manual inspection workflows, reducing client costs by 25%. Deployed via REST API on embedded hardware.
Focused on building AI that generalises beyond benchmarks — with rigor, reproducibility, and real deployment constraints in mind.
Applying deep learning to medical image analysis with a focus on high accuracy on low-cost, resource-constrained hardware. Work spans retinal disease classification with 85K+ image datasets.
Computer VisionUsing molecular fingerprints (Morgan, MACCS), ML models, and docking simulations to predict compound bioactivity. Focus on reproducible research with rigorous cross-validation.
LightGBM · RDKitDesigning real-time posture detection and correction systems for mobile hardware. Bridging a gap in mobile health tooling with patent-filed technology.
YOLO · TFLiteRemoving 5% noisy samples improved R² from 0.83→0.87 — data quality beats model complexity.
Rigorous cross-validated evaluation, Optuna hyperparameter search, and version-controlled pipelines.
Models validated on actual hardware — embedded systems, mobile — not just academic benchmarks.
Bisleri (Industry Collaboration)3 Months
Leading development of an automated computer vision–based quality inspection system for Bisleri bottles to detect plastic particles, external defects, and other bottle anomalies on a high-speed conveyor line.
Developed a two-stage vision pipeline combining monochrome imaging with 850 nm NIR illumination to detect particles inside transparent bottles, including particles in the >500 µm range.
Designed a controlled dark-tunnel imaging enclosure with optimized lighting, camera, lens, and sensor triggering to minimize reflections and false detections.
Developed custom image-processing software for automated particle detection and achieved ~98% detection accuracy with 1–2% false alarms during controlled testing.
Codeflow Studios6 Months
Developed a MobileNetV3 Large computer vision model for automated crack detection and severity grading on construction walls, enabling real-time field assessment directly on embedded hardware.
Replaced manual structural inspection workflows, reducing the client company's inspection costs by 25% and significantly cutting time-on-site per assessment.
Managed the full ML pipeline — data collection, augmentation, training, hyperparameter tuning, and hardware deployment — integrating the model into a production system via REST API.
Trusted Research Background
Bringing academic rigor and production-grade engineering to your AI project.
Start a Project→24-Hour Hackathon · AI in Healthcare
Secured 2nd place with Team Lakshya at MIT Railway Engineering College Barshi's 24-hour hackathon in the AI in Healthcare domain.
48-Hour Challenge · AI Domain · ₹50,000 Prize
After a grueling 48-hour challenge, Team Lakshya emerged victorious — winning the runner-up position and ₹50,000 prize with recognition on a national stage.
International · AI in Healthcare · ₹30,000 Prize
Won 1st place at PICT's International Hackathon in the AI in Healthcare domain, taking home a ₹30,000 prize.
National · AI & CyberSecurity Domain
Secured 3rd place at the Navonmesh National Hackathon at SSIPMT Raipur in the AI & CyberSecurity domain.
National · AI Domain · Trophy
Won 3rd prize with a trophy at the INFOTSAV National Hackathon hosted by IIIT Gwalior in the AI domain.
Inter-State · AI Domain
Secured 3rd place at the HackSprint Inter-State Level Hackathon in the AI domain.
International · 1200+ Participants
Selected as a finalist among 1200+ international participants at the Easy-A Consensus hackathon in Hong Kong.
Led technical execution across two active research collaborations — eye disease detection and Alzheimer's drug discovery — coordinating cross-disciplinary teams of 6–7 members each.
Represented teams across national and international competitions, achieving multiple podium finishes. Led team strategy, task delegation, and live execution under competition pressure.
Tutored 10+ students in AI and machine learning fundamentals. Designed and delivered structured learning paths covering Python, ML theory, and hands-on project work.
Organised workshops, coding sessions, and speaker events as a core member of the AI & Data Science Club at MMCOE. Helped grow student interest in applied AI.
I specialise in computer vision, cheminformatics, and applied ML — from medical diagnostics (retinal disease detection) to drug discovery (bioactivity prediction) to real-time posture detection on mobile. I focus on models that actually deploy on real hardware, not just benchmark scores.