TOOLKIT
PEEL. PLACE. EXPLORE.
Case Studies
Long AI Conversations – Solving Lag, Context & Continuity Challenges
Problem: Long AI conversations can become slow, laggy, or crash, with context degradation and hallucinated responses forcing users to rebuild context
Solution: A user-controlled workflow that converts conversations into structured, editable context document that enables seamless continuation across new conversations.
Includes: Problem Research • UX Design • User Flows • Wireframes • System Architecture • Key Interaction Decisions • Business Impact • Figma Prototype
Company-Specific AI – Role-Aware AI for Enterprise Workflows
Problem: Employees repeatedly explain company context, priorities and role-specific requirements before AI provide ranswers.
Solution: A Company Mode that combines shared organizational context with employee roles, enabling AI to reason automatically according to company and role-specific priorities.
Includes: Problem Research • UX Design • Enterprise Workflows • User Flows • Wireframes • System Architecture • Key Interaction Decisions • Business Impact • Figma Prototype
Projects
Adaptive Voice Assistant
TECH STACK:
Python • React • TypeScript • Vite • FastAPI • WebSocket • SQLite • Gemini • Chatterbox TTS • Docker
Developed an Adaptive Voice Assistant — a personalized, context-aware AI system enabling human-centric interactions through familiar voices, adaptive responses, and dual-mode (professional + personal) functionality. Selected among the Top 100 teams nationwide and Top 10 in Mumbai at the AMD Slingshot Hackathon.
Uber Rides– Exploratory Data Analysis (EDA)
TECH STACK:
Python • Pandas • NumPy • Seaborn • Matplotlib
Exploratory data analysis (EDA) of Uber ride data using Python, including data cleaning, feature engineering, and visualization to identify peak hours, demand trends across days and months, and trip purpose distribution.
Fleet Management Dashboard (PowerBI)
TECH STACK:
Power BI • DAX • Power Query • Excel/CSV • Data Modeling • Dashboard Design • KPI Analysis
Built a multi-page Fleet Analytics dashboard in Power BI analyzing fuel consumption, maintenance costs, driver-level performance, revenue trends, and geographic distribution, enabling cost optimization, operational efficiency, and customer retention insights.
HR Analytics Dashboard (PowerBI & Tableau)
TECH STACK:
Powerbi • Tableau • Excel/CSV
Developed interactive HR analytics dashboards in Power BI and Tableau to analyze employee demographics, performance, satisfaction, and attrition, delivering actionable insights into workforce trends and turnover patterns.
Financial Performance Dashboard (Tableau)
TECH STACK:
Tableau • Excel/CSV • LOD Expressions
Developed interactive HR analytics dashboards in Power BI and Tableau to analyze employee demographics, performance, satisfaction, and attrition, delivering actionable insights into workforce trends and turnover patterns.
NexGenTutor: Multi-Agent AI System for STEM Video Generation
TECH STACK:
Python • Gemini • LiteLLM • RAG (ChromaDB) • Manim • Streamlit • Kokoro TTS • FFmpeg • Async/Concurrent Processing • JSON-based configuration
Built NexGenTutor, an AI system that autonomously generates and evaluates STEM videos using LLM orchestration, RAG, Manim animation and TTS. Designed a multi-stage pipeline (planning → generation → rendering → composition) with a Streamlit UI for scene previews and control. Architected for multi-LLM support (currently using Gemini) with modular, scalable workflows and concurrent scene generation.
Chat with Documents: API-Free, Low-Compute Meta AI System-
TECH STACK:
Streamlit • LangChain • FAISS • Sentence Transformers • Meta AI • PyPDF2 • python-docx • BeautifulSoup • Tesseract OCR
Streamlit-based document Q&A system supporting PDFs, DOCX, HTML and images, featuring OCR, semantic chunking and FAISS-based indexing, with lightweight Meta/LLaMA-style inference for low-compute, keyless deployment.
Comparative Analysis of Symbolic Music-Generation Models
TECH STACK:
Python • TensorFlow/Keras • NumPy • music21 • librosa • matplotlib • Deep Learning
Built an symbolic music generation and benchmarking pipeline using the MIT Irish Folk Music Dataset (ABC notation).Trained and compared RNN, LSTM, GRU, CNN, Transformer, VAE and GAN models using a unified evaluation framework spanning model-level metrics (loss, accuracy, perplexity and GAN-specific metrics), structural metrics (BLEU, n-gram overlap, Jaccard overlap, sequence similarity, melodic contour,pitch-class histogram divergence, musical coherence analysis), confusion matrix and audio similarity using DTW.
Ai Resume Analyzer
TECH STACK:
Python • Google Gemini (Gemini Pro Vision) • Streamlit • pdf2image • PIL • python-dotenv
Built an LLM-powered ATS resume analyzer using Google Gemini (Gemini Pro Vision), implementing multimodal resume–job description matching to generate match scores, keyword gap analysis, resume summaries, strengths and weaknesses, and actionable role-fit insights for candidates to optimize resumes and for recruiters to identify the right candidates.
AyurCare: AI-Based Disease Detection Web Platform
TECH STACK:
Python • Flask • TensorFlow (CNN, VGG16) • Scikit-learn (Random Forest, SVM) • XGBoost • OpenCV • Imutils • HTML • CSS • Anaconda
AI-powered healthcare web platform integrating four disease detection models using ML/DL techniques, deployed via Flask for real-time predictions. Implemented VGG16-based CNN for Brain Tumor detection (92.42%), Random Forest for Breast Cancer (99.53%), SVM for Diabetes (78.66%) and a custom CNN for Pneumonia Detection (98.16%), demonstrating an end-to-end pipeline from model training to web deployment.
Inventory Management System (IMS): Institute-level project
TECH STACK:
PHP • SQL • HTML • CSS • Javascript
Developed a centralized web-based inventory management system for college laboratories using PHP and MySQL, enabling efficient tracking of equipment lifecycle, cost, and status. Implemented secure authentication with role-based access control (e.g., lab in-charge, admin) and modular features including hardware/software inventory management, item addition, borrower tracking, repair and defect handling, scrap management, user profile management, and an administrative dashboard. Improved asset visibility and streamlined coordination across multiple labs.
Research
Self-Updating Multi-Agent Cognitive Architecture for Safe AGI
Brain-inspired LoRA-based multi-agent cognitive architecture where a shared LLM acts as the central brain and specialized agents function like different brain regions for reasoning, memory, learning, safety, and execution. The system enables autonomous learning, self-extension, continuous adaptation, and human-governed model evolution through orchestrated agent communication and modular LoRA updates as a scalable pathway toward safe AGI.
Comparative Analysis of Symbolic Music-Generation Models
Conducted a research-driven comparative study of deep learning architectures for monophonic music generation using the MIT Irish Folk Music Dataset (ABC notation). Implemented and benchmarked RNN, LSTM, GRU, CNN, Transformer, VAE and GAN models using a unified evaluation framework metrics. The study focused on fairly comparing architectures rather than maximizing individual model performance, with all models trained using the same hyperparameters to ensure unbiased evaluation and identify promising architectures for future music-generation systems. See the corresponding project in the Projects section for implementation details and results.
AI for Healthcare Biotechnology — IJSREM
Authored and published a survey-based research paper exploring the applications of Artificial Intelligence in healthcare and biotechnology, with emphasis on medical diagnostics, drug discovery, personalized medicine, clinical decision support, and AI-driven healthcare systems. The research also analyzed ethical challenges, data privacy concerns, and future trends in AI-powered healthcare technologies. Presented at the National Conference on AI Revolutions–2023 and published in IJSREM, Volume 08, Issue 02 (February 2024).