Hi Contra community!
Just wrapped a rewarding project for Nauvyashree—delivered deep-dive EDA, custom visualizations, and rich survey sentiment analysis across four datasets. Each milestone included well-designed PDFs packed with clear charts and approachable explanations—making insights accessible for all audiences.
I went the extra mile with collaborative Google Meet sessions to walk the client through every analysis, helping connect the findings to big-picture goals. Feedback was fantastic, and milestone-based payments made the process smooth and transparent.
This project really strengthened my ability to turn complex results into stories that empower client decision-making. If you want visually engaging, client-focused analytics—or need easy-to-follow reports and hands-on walkthroughs for your next project—let’s connect!
CaratIQ — Modern Jewelry Management & Catalog Platform
Managing dynamic jewelry inventories across multiple sales channels often breaks under traditional database models due to real-time pricing recalculations and complex SKU variations.
What I Built:
• Dynamic Catalog & Multi-Attribute Filtering: Architected sub-150ms catalog searches handling live metal spot pricing, carat grades, and stone certification tiers without client-side lag.
• Server-Rendered Architecture: Leveraged Next.js Server Components to offload bulk queries, significantly reducing initial client bundle size.
• Automated Inventory & Invoicing Sync: Integrated custom webhooks and payment flows for instant stock reservation and order processing.
• Role-Based Access Control: Built multi-tenant admin dashboards for floor staff, store managers, and suppliers.
Tech Stack:
Next.js, React, Node.js, PostgreSQL, Tailwind CSS, REST APIs & Webhooks
Results:
• Sub-150ms Query Speeds: Cut search and filtering latency by over 40% across extensive catalog datasets.
• Zero Sync Collisions: Automated live price updates and inventory locks, eliminating overselling across peak sales windows.
The Outcome:
Transformed a traditionally fragmented retail inventory system into a unified, high-speed SaaS platform that bridges in-store sales and digital catalog operations.
National Accident Statistics Dashboard in Google Looker Studio:
I built an interactive Looker Studio dashboard using Google Sheets as the data source to visualize national accident statistics. The report analyzes accidents by region, time, severity, and demographics. A scheduled data re:fresh every 15 minutes keeps the dashboard updated in real time. As a Data Visualization Expert, I combined automation and design to deliver a clean, insightful, and dynamic Looker Studio dashboard for data-driven decision-making.
🎉 💯 🎉
Aspiring dev crafting practical, reliable systems.
Aspiring dev crafting practical, reliable systems.
Deepfake Detection System
Problem: Deepfake accessibility creates real risks—misinformation, identity fraud, digital trust erosion. I built a production-grade detector: fast, accurate, explainable, CPU-deployable.
Approach: Transfer learning with ResNet-50 (25.5M params, pre-trained ImageNet-1K V2). Rather than training from scratch on limited data, I leveraged rich hierarchical features that transfer exceptionally well to detecting artifact patterns (blending boundaries, color mismatches, compression artifacts).
Key Decisions:
Face-Centric Preprocessing: OpenCV DNN crops faces to 224×224, concentrating attention on artifact-rich regions (eyes, mouth, jawlines).
Custom Classification Head: Bottleneck design (2048→512→1) with Focal Loss, label smoothing, progressive unfreezing—prevents overfitting on ~4,000 Ciplab images.
Explainable AI: Custom Grad-CAM generates attention heatmaps showing exactly where the model detected artifacts.
Real-World Robustness: Trained on JPEG compression, Gaussian noise, blur augmentations to handle compressed video feeds.
Results:
AUC-ROC: 0.9424 | Accuracy: 87.25% | Precision/Recall: 87.27%/84.81%
Inference: <100ms per face on CPU
Live on Hugging Face Spaces (no GPU)
Tech: PyTorch, OpenCV, Albumentations, Gradio. Images, videos, live webcam—one codebase.
Demonstrates: End-to-end AI engineering with deliberate trade-offs (efficiency over raw accuracy), training discipline (Focal Loss, Cosine Annealing), and production-first thinking (CPU compatibility, interpretability, real-world robustness).
Demo: https://huggingface.co/spaces/Shri04/deepfake-detector
Government School ERP System — Full-Stack Platform for a Real Government School Client
Designed and built a bilingual, full-stack ERP system for a government school in Uttar Pradesh — replacing paper attendance registers, handwritten admission forms, and manual Mid-Day Meal logs with a secure, role-based digital platform.
Architected the full data layer — a 25+ model relational schema (PostgreSQL + Prisma) mapped directly from the client's actual government paperwork, including India's UDISE+ admission forms and Form-34 Transfer Certificates, not a generic template.
Built defense-in-depth security — role-based access control enforced at the server layer (not just route guards), brute-force account lockout, and a login system designed around a real constraint: students without personal phones authenticate by admission number instead.
Shipped a bilingual UI from day one — Hindi and English throughout, using next-intl, so the interface actually works for the people using it, not just the developer testing it.
Automated the admission pipeline end-to-end — a 30+ field digital intake form (photo + document uploads to cloud storage) that auto-generates student accounts and credentials in a single atomic transaction.
Still actively building — Mid-Day Meal tracking and automated report card generation are next. Open to freelance and contract work where someone needs to own both the system architecture and the day-to-day execution. If it's real, complex, and needs to actually ship — let's talk.
1
164
Full-Stack AI-Powered Job Search Platform | Next.js · TypeScript · PostgreSQL · Groq AI · n8n
Built a production-ready web application to help job seekers manage their entire job hunt in one place — replacing messy spreadsheets with an intelligent, automated pipeline.
Developed a drag-and-drop Kanban board tracking applications across 6 stages (Applied → Offer → Rejected) with real-time updates
Integrated Groq LLaMA 3.3 70B via n8n automation workflows to power an AI chat assistant that lets users query and update their pipeline conversationally
Built an ATS Resume Scorer — paste any JD to get a 0–100 match score, keyword gap analysis, strengths/gaps breakdown, and interview likelihood rating
Built an AI Resume Tailor with two modes: section-by-section suggestions and a full ATS-optimised rewrite — output downloadable as .txt
Implemented Google OAuth auth, serverless PostgreSQL (Neon), PDF resume parsing, and JD auto-extraction — all at $0/month infrastructure cost
Stack: Next.js 16 · TypeScript · Prisma · PostgreSQL (Neon) · Groq API · n8n · NextAuth.js · Docker