A full-stack, AI-powered web vulnerability scanner built with Streamlit and powered by a dual-intelligence engine: Groq Cloud AI (LLaMA 3.3 70B) for contextual threat analysis and a trained Random Forest ML model for pattern-based detection. Users authenticate via a login system, enter any URL or domain, and the scanner automatically discovers subdomains, crawls pages, and surfaces vulnerabilities across SQL Injection, XSS, CSRF, insecure headers, and security misconfigurations — all exported as JSON, Markdown, or CSV reports.

Manual penetration testing is slow, expensive, and hard to scale. Rule-based scanners miss context-dependent vulnerabilities and generate noisy false positives. This project combines a trained ML model with a real Groq LLM to create a scanner that understands page context, validates findings intelligently, and produces executive-grade reports — all runnable with a single streamlit run app_ai.py command.

• Layer 1 — Random Forest ML Model: A trained scikit-learn Random Forest classifier predicts SQL Injection, XSS, or Misconfiguration with per-class probability scores. Only predictions above 60% confidence are surfaced as findings
• Layer 2 — Traditional Pattern Detection: Regex and DOM heuristics scan page HTML for known dangerous patterns — eval(), innerHTML=, document.write(), GET-method forms, POST forms without CSRF tokens, and sensitive keywords in HTML comments
• Layer 3 — Groq AI Analysis (LLaMA 3.3 70B): The GroqOrchestrator sends HTML snippets, form structures, and script counts to Groq's API for deep contextual analysis, returning structured JSON vulnerability objects with CWE IDs, proof of concept, and remediation guidance

• Domain Recognition: Uses LLaMA to intelligently parse any user input — full URLs, bare domains, subdomains, paths — and extract a clean domain with confidence score
• Page Content Analysis: Sends truncated HTML + form/script metadata to Groq, receiving a JSON array of vulnerabilities with risk scores calculated by severity × exploitability × business impact × confidence
• Vulnerability Validation: Each detected finding is sent back to Groq for a second-pass review, refining severity and filtering false positives
• Executive Summary: Generates a 200–250 word business-readable summary with security posture rating, key findings, and immediate action items

• crt.sh Certificate Transparency Logs + HackerTarget API + 20-word common subdomain wordlist
• 30-thread ThreadPoolExecutor for parallel DNS resolution and HTTP reachability verification
• Smart Crawl mode prioritises security-relevant pages: login, admin, upload, search, payment, api, profile
• Configurable scan depth (Quick / Standard / Deep / Comprehensive) with 5–30 pages per domain slider

• SQL Injection (CWE-89), XSS (CWE-79), Missing CSRF Protection (CWE-352)
• Sensitive Data Exposure (CWE-200) via HTML comment scanning
• Insecure Protocol (CWE-319) and five missing security headers (CWE-16)
• ML-detected Misconfiguration based on CSP absence and script/form count thresholds

Python, Streamlit, Groq Cloud API (LLaMA 3.3 70B), Scikit-learn (Random Forest), Joblib, Pandas, NumPy, BeautifulSoup4, Requests, python-dotenv, lxml, concurrent.futures, socket

A machine learning-powered system designed to detect malicious network activity in real time, providing proactive cybersecurity through traffic pattern analysis and threat classification.

Modern networks face threats including malware, DoS attacks, and unauthorized access. Traditional rule-based tools struggle with unknown threats. This project applies ML to recognize attack patterns in network traffic.

Random Forest, SVM, and ensemble learners classify traffic as benign or malicious. Models are evaluated using accuracy, precision, recall, and F1-score metrics.

• Network traffic preprocessing and feature preparation
• Real-time inference through Flask API
• Confidence scores for predictions
• Attack category visualisation

Enhances traditional security tools by detecting complex attack patterns, supporting faster incident response and reducing false positives. Achieved 94% detection accuracy during the RedKross internship deployment.

A real-time music recommendation system built using Python and the Spotify Web API. Its goal is to help users discover new music tailored to their listening preferences by analyzing their historical listening behavior and track audio features.

With millions of songs available on streaming platforms, users often struggle to find tracks that match their tastes. This project creates a system that recommends songs dynamically based on listening patterns and audio characteristics.

• User Authentication: Secure login via Spotify OAuth 2.0
• Data Retrieval: Collects recently played tracks and audio features
• Feature Processing: Normalizes song vectors for similarity comparison
• Similarity Calculation: Uses cosine similarity for matching
• Recommendations: Real-time suggestions through web interface

• Real-time analysis enables highly personalized recommendations
• Audio feature-based similarity outperforms playlist matching
• API optimization significantly improves responsiveness

Python, Flask, Pandas, NumPy, Spotify Web API

ML-powered car price prediction system estimating vehicle values based on mileage, manufacture year, brand, and other features through regression models.

Predicting fair market value for used cars is challenging. Buyers and sellers rely on rough estimates. This model learns from historical data to provide accurate predictions.

Data cleaning handled missing values and categorical encoding. Feature engineering created variables like vehicle age, enhancing the model's pattern recognition.

Regression models trained on historical car data. Evaluation using MAE and R² score ensures generalisation to unseen data.

• Data preprocessing and transformation pipeline
• Real-time prediction via Flask API
• Cloud deployment ready
• Serialised model for quick inference

A Python-based recommendation system providing personalized movie suggestions using content similarity algorithms and collaborative filtering.

Users struggle to find relevant movies among thousands of titles. Static charts don't cater to individual tastes. This system analyses metadata and similarity patterns for personalisation.

Precomputed similarity matrices enable fast recommendations. Cosine similarity between movie feature vectors (genres, ratings) drives the logic.

• Interactive web interface
• Fast real-time recommendations
• Personalised suggestions
• Lightweight Flask deployment

Enhances content discovery and user engagement by applying ML techniques for personalized recommendations in large catalogs.

In-depth exploratory analysis of IPL cricket (2008–2019), uncovering patterns in team performance, player contributions, and match outcomes using historical data.

IPL generates massive datasets, but raw statistics don't explain performance trends. This project structures data to answer questions about consistency, toss influence, and venue effects.

• Team Performance: Win percentages and dominance patterns
• Toss Impact: Correlation with match outcomes
• Player Analysis: Strike rates and consistency metrics
• Venue Insights: Stadium influence on scoring

• Small group of teams showed long-term dominance
• Toss had limited impact on outcomes
• Individual performance influenced success more than conditions
• Venue characteristics affected scoring trends