KT
PORTFOLIO INDEX
KRISH TRIVEDI · AHMEDABAD, GUJARAT, INDIA
04 — PORTFOLIO & ARCHITECTURAL SHOWCASE

Projects & Deployments

Real-world applications, live production cloud platforms, and machine learning models engineered by Krish Trivedi. Each project includes complete problem statements, architectural solutions, and verified codebases.

KisanMitra AI Krishi Officer Screenshot Preview
SIH 2025 NOMINEE
Status: Hackathon Finalist Dataset: PlantVillage & Kaggle
PROJECT 01 // AGRI-TECH AI | Computer Vision + LLM

KisanMitra – AI Krishi Officer

Multilingual Crop Pathology & Voice-Enabled Advisory Assistant
Problem Statement: Rural farmers lack accessible agronomy experts when crops develop fungal or pest blight, leading to high crop yield losses and pesticide misapplication.
Project Objective: Deliver instant leaf disease identification from mobile camera uploads and voice advisories spoken in Gujarati and Hindi for maximum accessibility.
Solution Overview:

Constructed an end-to-end Python Flask inference pipeline connecting a convolutional neural network (PyTorch/ResNet) crop disease classifier with OpenAI API dosage generation. Results are converted into localized Gujarati and Hindi speech prompts.

Key Features:
Foliage image disease scan (>90% conf)
Gujarati & Hindi speech synthesis
Precise chemical dosage calculation
Outbreak mapping & SQLite history
Technology Stack:
Python Flask PyTorch / CNN OpenAI API Speech Synthesis SQLite
My Contribution: Lead AI Developer for SIH 2025; engineered computer vision model pipeline, prompt safety rules, and Flask backend.
HPAD Travel Management System Live Screenshot
LIVE ON RENDER
Status: Production Live Runtime: Node.js 20+ (Next.js 15)
PROJECT 02 // ENTERPRISE PLATFORM | Full-Stack Cloud App

HPAD Travel Management System

Digital Employee Travel & Site Visit Operations Platform
Problem Statement: Corporate travel coordination relied on paper trails, emails, and static spreadsheets, causing booking delays and lack of visibility into engineer site movements.
Project Objective: Digitalize travel requests, flight booking coordination, approval hierarchies, and expense tracking into a secure, role-based cloud platform.
Solution Overview:

Built with Next.js 15 App Router, React 19, and Supabase (PostgreSQL). Provides granular Role-Based Access Control (Employee, Admin, Super Admin), real-time Recharts dashboards, and automated PDF/Excel movement reports.

Key Features:
Role-Based Access Control (RBAC)
Recharts analytics visualization
Automated Excel reporting (ExcelJS)
PDF travel vouchers via jsPDF
Technology Stack:
Next.js 15 React 19 TypeScript Supabase (PostgreSQL) Tailwind CSS v4 Render Cloud
My Contribution: Full-stack software architecture, schema design, authentication middleware, PDF/Excel export routines, and Render deployment.
White Swan Holidays Website Screenshot Preview
LIVE ON RENDER
Status: Production Live Hosting: Render Static Service
PROJECT 03 // TRAVEL COMMERCE | Multi-Page Web Platform

White Swan Holidays

Tourism Portal with Tailor-Made Itineraries & WhatsApp Lead Dispatch
Problem Statement: High form abandonment and difficulty browsing complex domestic and international holiday packages on mobile browsers.
Project Objective: Create a high-speed, mobile-responsive portal with WhatsApp lead integration and Schema.org SEO for high search visibility.
Solution Overview:

Structured multi-page architecture separating Domestic, International, MICE, and Visa services. Integrated 1-click WhatsApp business lead dispatch and deployed on Render with Cloudflare edge optimization.

Key Features:
Domestic & International packages
WhatsApp Business 1-Click Consultation
Schema.org TravelAgency JSON-LD metadata
Continuous Render cloud auto-deploy
Technology Stack:
HTML5 Semantic CSS3 Custom JavaScript (ES6+) Schema.org JSON-LD Cloudflare CDN Render
My Contribution: Front-end web development, responsive CSS styling, WhatsApp integration, and Render CI/CD deployment.
Loan Default Prediction ROC-AUC Curve Mockup
SCIKIT-LEARN ML
Status: Model Trained & Validated ROC-AUC: 0.884
PROJECT 04 // MACHINE LEARNING | Predictive Analytics

Loan Default Prediction Model

Supervised Learning Credit Risk Classification Pipeline
Problem Statement: Financial lending institutions incur significant losses due to inaccurate borrower credit risk classification and imbalanced historical loan datasets.
Project Objective: Develop a predictive classification pipeline that preprocesses multi-attribute financial records and forecasts the probability of loan default with high precision.
Solution Overview:

Applied Exploratory Data Analysis (EDA), feature engineering, KNN missing-value imputation, and SMOTE resampling. Evaluated Random Forest, Logistic Regression, and Decision Trees using 5-Fold Stratified Cross Validation.

Key Features:
ROC-AUC 0.884 performance
DTI & Revolving Credit Gini rankings
SMOTE class imbalance mitigation
Confusion matrix metric reporting
Technology Stack:
Python Pandas NumPy Scikit-learn Matplotlib / Seaborn
My Contribution: End-to-end data pipeline, feature scaling, model training, cross-validation, and metrics analysis.
AI Psychology Chatbot Session Mockup
CONVERSATIONAL AI
Status: Experimental Prototype Guardrails: 100% Non-Clinical
PROJECT 05 // CONVERSATIONAL AGENT | NLP & Prompt Safety

AI Psychology & Wellness Chatbot

Empathetic Mental Wellness Companion with Ethical Boundary Guardrails
Problem Statement: Students and young professionals often experience episodic stress and lack access to immediate, non-judgmental reflective dialogue tools.
Project Objective: Develop an ethical conversational assistant that practices active listening, guides reflective journaling, and implements strict non-clinical safety boundaries.
Solution Overview:

Engineered using Python and OpenAI prompt architecture. Includes emotional polarity detection, multi-turn conversational memory, and hardcoded refusal guardrails prohibiting medical diagnosis or psychiatric advice.

Key Features:
Contextual conversational memory
Strict non-clinical ethical guardrails
Sentiment polarity detection trigger
Structured cognitive de-escalation
Technology Stack:
Python OpenAI API NLP Prompt Engineering JSON Session Store
My Contribution: System prompt architecture, boundary safety enforcement, and sentiment analysis integration.
Web Scraping Stream Mockup
DATA EXTRACTION ETL
Status: Production Pipeline Throughput: ~418 records / sec
PROJECT 06 // DATA PIPELINE | Automated ETL

Web Scraping & Data Aggregation Pipeline

High-Throughput Asynchronous Extraction & Sanitization Engine
Problem Statement: Manually compiling commercial directory and market intelligence data across fragmented websites is tedious, error-prone, and unsustainable.
Project Objective: Develop an asynchronous, resilient scraping pipeline that harvests thousands of structured records per hour while respecting rate limits and handling schema drift.
Solution Overview:

Python architecture using Asyncio, Aiohttp, and BeautifulSoup. Includes resilient retry logic, automated DOM sanitization, deduplication hashing, and export pipelines to CSV, JSON, and Parquet.

Key Features:
16 parallel worker threads
Automated DOM entity sanitization
Duplicate elimination hashing
Direct export to CSV, JSON & Parquet
Technology Stack:
Python 3 Asyncio / Aiohttp BeautifulSoup4 Pandas JSON / Parquet
My Contribution: Thread pooling architecture, rate limiter implementation, and data sanitization routines.
GITHUB API SYNC

Live GitHub Repositories

github.com/Krishhtrivedi
Syncing verified public repositories...
← 03 // AI Laboratory 05 // Skills & Technologies