AI/ML Engineer with a proven ML internship at SaiKet Systems, 6+ shipped projects (NLP, GenAI, Computer Vision), and 150+ LeetCode problems solved. Building intelligent applications that deliver real business impact — currently targeting 10 LPA+ roles.
Worked as a Machine Learning Intern at SaiKet Systems (Aug – Sep 2025), demonstrating exceptional analytical thinking, data modeling skills, and effective communication abilities. Displayed remarkable dedication and passion for Machine Learning throughout the internship.
Intensive live training on C++ Programming: OOPs and DSA — 35+ hours covering Object Oriented Programming principles and Data Structures & Algorithms fundamentals.
C++ ProgrammingOOPsData Structures & Algorithms35+ Hours
Jun – Jul 2025✓ Completed
✓ VERIFIED
My Achievements
Competitions, challenges and participation milestones.
🏅 Participation
Internship Common Aptitude Test (iCAT)
Participated in the national-level Internship Common Aptitude Test, demonstrating aptitude and commitment to professional growth.
iCATMar 2026
🚩 CTF Challenge
Hack Quest – 24 Hours CTF Challenge
Participated in Concoction 2024 — Intra University Tech Fusion, a 24-hour Capture The Flag cybersecurity challenge at LPU in collaboration with upGrad Campus.
upGrad Campus × LPUApr 2024
CREDENTIALS
My Certifications
Verified credentials from HackerRank, Infosys, Coursera, LinkedIn & more.
🔍 Just built a Fake Social Media Account Detector using ML!
Engineered 15+ features from profile metadata and trained Random Forest, SVM & Logistic Regression. Achieved 92% precision — excited to keep building real-world ML applications!
SD
Sonika Deshwal
AI/ML Engineer · Software Developer
2 months ago
🎤 Project
♟️ Built a Voice-Controlled Chess AI — and it works!
Say "knight to f3" and it moves! 94% recognition accuracy with <500ms latency. Built with Python, SpeechRecognition and a custom NLP parsing layer with fuzzy matching. So proud of this one!
SD
Sonika Deshwal
AI/ML Engineer · Software Developer
3 months ago
📊 Achievement
🎉 Earned 5-Star C++ on HackerRank!
Consistent practice on Data Structures and Algorithms finally paid off. From arrays to graphs — every problem solved has built my confidence. Next target: LeetCode 200+ problems!
SD
Sonika Deshwal
AI/ML Engineer · Software Developer
4 months ago
💡 Learning
✉️ Smart Email Classifier — 89% accuracy with Naive Bayes!
Built an NLP pipeline that automatically categorizes emails into spam, promotion, important and urgent using TF-IDF. 5-fold cross-validation gave me confidence in the results. Loved every step of this!
SD
Sonika Deshwal
AI/ML Engineer · Software Developer
5 months ago
🎓 Certification
📜 Completed ChatGPT-4 Prompt Engineering by Infosys & Springboard!
Generative AI is the future and I want to be ready for it. This course gave me hands-on experience with LLMs, prompt design patterns and responsible AI. Excited to apply this in my next project!
SD
Sonika Deshwal
AI/ML Engineer · Software Developer
2 weeks ago
🚀 Project Launch
🎬 Introducing DeepNote AI — Turn any YouTube video into structured notes!
Powered by Groq LLaMA 3.3 70B, DeepNote AI extracts timestamps, key insights and generates RAG-ready knowledge from any video. Built with Streamlit + Python. This one's special 💙
SD
Sonika Deshwal
AI/ML Engineer · Software Developer
1 month ago
🤖 AI Product
🎤 Smart AI Interview Coach is LIVE — practice smarter, not harder!
Built a full AI mock interview platform: LLM scoring, voice Q&A, resume-based question generation & downloadable PDF reports. 300+ questions across DSA, ML, and system design. Try it!
SD
Sonika Deshwal
AI/ML Engineer · Software Developer
6 weeks ago
📜 Certification
🏅 Certified in C++ OOP & DSA — 35+ hours of grinding paid off!
Completed an intensive 35-hour C++ certification covering OOP, STL, memory management and competitive DSA. Immediately applied it to solve 150+ LeetCode problems. The grind never stops 💪
SD
Sonika Deshwal
AI/ML Engineer · Software Developer
2 months ago
🏆 Achievement
🚩 Participated in Hack Quest — 24hr CTF at LPU × upGrad Campus!
Spent 24 non-stop hours cracking cybersecurity challenges at Concoction 2024. Flag captured, lessons learned, friendships forged. This is what tech culture looks like 🔐🔥
Built an interactive data analytics dashboard to analyze sales, profit trends, customer behavior, and regional performance. Performed complete data cleaning, preprocessing, and transformation using Pandas.
Key Features
Dynamic filtering by date, category, and region enabling real-time KPI analysis
Interactive visualizations for sales trends, profit margins, and customer segmentation
Data cleaning pipeline handling missing values and outliers
Regional performance heatmaps and product-level drill-down analysis
Deployed as a Streamlit web application for easy browser access
Technical Approach
Used Pandas for data wrangling, NumPy for numerical computations, and Matplotlib/Seaborn for visualizations. Streamlit was used to build the interactive web dashboard with live filtering capabilities.
Automated NLP pipeline categorizing 50+ daily emails into spam, promotion, important, and urgent using TF-IDF vectorization and Naive Bayes classifier. Deployed with real-time prediction capability.
Key Features
89% classification accuracy with 5-fold cross-validation
TF-IDF vectorization for feature extraction from raw email text
Multi-class: spam, promotion, important, urgent
Real-time prediction pipeline for new incoming emails
Confusion matrix and precision-recall analysis for evaluation
Technical Approach
Applied NLP preprocessing (tokenization, stopword removal, stemming) before TF-IDF vectorization. Trained Naive Bayes classifier with hyperparameter tuning via cross-validation.
Real-time voice-controlled chess converting speech like "knight to f3" into legal moves with 94% recognition accuracy. Features NLP parsing with fuzzy matching and complete rule validation.
Key Features
94% speech recognition accuracy for chess move commands
NLP parsing with fuzzy matching for natural language input
Chess rule validation ensuring only legal moves are accepted
Modular OOP architecture for scalability
<500ms end-to-end latency from voice to board update
Technical Approach
Built using SpeechRecognition and PyAudio for audio input. Custom NLP layer parses chess notation from natural speech, integrated with Python-Chess for board state management.
End-to-end supervised ML pipeline detecting fake accounts using profile metadata and behavioral features. Achieved 92% precision and 89% F1-score across Logistic Regression, Random Forest, and SVM.
Key Features
Engineered 15+ features from profile metadata and behavioral data
Compared Logistic Regression, Random Forest, and SVM
92% precision and 89% F1-score on test set
GridSearchCV for systematic hyperparameter optimization
Extensive feature engineering on profile metadata with SMOTE for class imbalance. GridSearchCV hyperparameter tuning — Random Forest achieved best results.
An AI-powered mock interview platform that generates personalized, role-based questions from your uploaded resume and evaluates every answer in real-time across 5 NLP dimensions — clarity, technical accuracy, communication, depth, and confidence — powered by Groq LLaMA 3.3 70B.
Key Features
Resume-driven question generation (PDF/TXT upload) — auto-tailored to your background
5-dimension AI scoring: Overall, Clarity, Technical, Communication, Depth
Voice interview mode: Speech-to-Text via mic + Text-to-Speech question read-aloud (gTTS)
One-click PDF interview report with per-question breakdown, keyword coverage & model answers
Built on Groq's ultra-fast LLaMA 3.3 70B inference for real-time answer evaluation. SpeechRecognition + PyAudio for live mic input with file-upload fallback. ReportLab generates professional PDF reports. Streamlit handles the full interactive UI with session state management.
Powered by Groq LLaMA 3.3 70B · RAG-Ready · More Advanced Than ScreenApp
StreamlitGroqLLaMA 3.3 70BRAGPythonNLP
ABOUT THE PROJECT
DeepNote AI transforms any video transcript into structured, actionable knowledge instantly. Paste a transcript and get organized notes, timestamped sections, action items, and key insights — powered by Groq's ultra-fast LLaMA 3.3 70B inference engine with RAG capabilities.
KEY HIGHLIGHTS
Converts raw video transcripts into structured notes with timestamps and action items
Powered by Groq LLaMA 3.3 70B — faster and more advanced than ScreenApp
RAG-Ready architecture for context-aware knowledge retrieval
Built with Streamlit for a clean, interactive web interface
Supports multiple video sources with real-time processing
