Discover a diverse portfolio of AI, ML, and DL projects, delivering advanced solutions in data analytics, automation, and intelligent system design
Master your journey with TrekSathi: The ultimate all-in-one travel ecosystem. Experience AI-driven itineraries, real-time budgeting, and an integrated marketplace for a seamless, smarter way to explore.
My journey building Recaply-AI: A mobile prototype that turns complex classroom lectures into instant, structured study notes using Groq and Gemini AI.
Abstract As Large Language Models (LLMs) evolve from simple text generators to reasoning engines, the focus of AI development has shifted toward agentic workflows—systems capable of autonomous planning, tool use, and self-correction. To explore the efficacy of modern orchestration frameworks, I engineered the Deep Research Agent: a fully autonomous system designed to perform iterative, multi-step research tasks. This project demonstrates a production-ready implementation of a cyclic graph architecture (LangGraph) utilizing the Groq API for high-throughput inference. The resulting system achieves professional-grade research synthesis with a marginal operating cost of $0.005 per query, proving that high-performance autonomous agents can be built cost-effectively using open-weight models. 1. System Overview: The Deep Research Agent The Deep Research Agent is not merely a wrapper for an LLM; it is a stateful application that mimics the workflow of a human analyst. Unlike zero-shot querying, this system employs an iterative “thought-loop” to refine information quality before generating a final response. Core Capabilities: 2. Technical Stack & Design Choices The architecture was chosen to maximize architectural flexibility while minimizing inference latency and operational costs. 3. Architectural Analysis: Cyclic Graph vs. Linear Chains A key engineering decision in this project was the implementation of a Cyclic Graph architecture over a traditional Linear Chain. 4. Engineering Implementation & Challenges The development process highlighted several critical aspects of building production-grade agents. A. State Management Implementation Effective state management is the backbone of any agentic system. I implemented a TypedDict structure with reducer operators to maintain context across iterations. This ensures that research findings are accumulated rather than overwritten during loops. Python B. Resilience and Error Handling To ensure robustness suitable for automated tasks, I implemented exponential backoff strategies for all external API calls. This prevents cascade failures during momentary latency spikes from search or LLM providers. Python C. Resource Optimization (Cost Analysis) A primary objective was to demonstrate the economic feasibility of running autonomous agents at scale. By optimizing the system prompt and pruning search results (limiting context window usage), the system achieves a 95% cost reduction compared to proprietary model APIs (e.g., GPT-4). Metric Standard API approach Deep Research Agent (Optimized) Cost Per Query ~$0.10 **~$0.005** Latency Variable < 3s (Inference) Architecture Black Box Open / Customizable 5. Conclusion & Future Scope This project validates that professional-grade AI agents do not require prohibitive budgets or closed ecosystems. By leveraging LangGraph for sophisticated orchestration and Groq for high-speed inference, I have engineered a system that is both autonomous and economically scalable. Future Research Directions: Repository: github.com/kazisalon/Deep-Research-Agent
A professional virtual trading simulator for the Nepal Stock Exchange. Practice NEPSE investing risk-free with real-time data and portfolio tracking
In a world where time seems to slip through our fingers, Live Dots offers a unique and beautiful way to visualize your year’s progress right on your phone’s wallpaper. This innovative Android live wallpaper app turns each day of the year into a visual dot, creating a stunning calendar that automatically updates daily to keep you mindful of time’s passage. What Makes Live Dots Special? Live Dots is more than just a wallpaper—it’s a daily reminder to make every moment count. The app displays a minimalist grid of dots representing every single day of the year, with each dot telling a story about where you are in your annual journey. Automatic Daily Updates The standout feature of Live Dots is its intelligent automatic update system. Once you set your wallpaper, the app works silently in the background to refresh your wallpaper once per day at midnight. This means: The app uses Android’s WorkManager to schedule these daily updates efficiently, ensuring your calendar stays current without draining your battery or requiring constant app launches. Stunning Visual Design Minimalist Dot Grid Layout Live Dots presents your year as an elegant grid of 365 dots (or 366 for leap years), arranged in a clean 15-column by 25-row layout. Each dot represents a single day: White dots – Days you’ve already lived this year Accent-colored dot – Today (the current day) Dark gray dots – Days yet to come This simple yet powerful visualization lets you see at a glance how much of the year has passed and how many days remain. How It Works 1. Install and launch Live Dots 2.Choose your accent color from four beautiful options 3.Preview your wallpaper to see how it looks 4.Apply to home screen, lock screen, or both 5.Confirm automatic daily updates 6.Relax – Your wallpaper now updates automatically every day at midnight! The Philosophy Behind Live Dots Time is our most precious resource, yet it’s easy to lose track of days, weeks, and months. Live Dots was created to help you: –Visualize time’s passage in a tangible way –Stay present and mindful of each day –Appreciate the time you have –Motivate yourself to make each day count Every time you unlock your phone, you’ll see a beautiful reminder of where you are in your year’s journey—not to stress you out, but to inspire you to live intentionally. Conclusion Live Dots is more than a wallpaper app—it’s a daily companion that helps you stay connected to the rhythm of your year. With its automatic daily updates, stunning visual design, customizable colors, and privacy-first approach, it’s the perfect blend of beauty and functionality. Transform your phone screen into a meaningful year tracker. Download Live Dots today and make every day visible.
Generative Adversarial Networks (GANs) have revolutionized image synthesis. In this post, we walk through the implementation of a Deep Convolutional GAN (DCGAN) using Keras and TensorFlow, trained to generate 64×64 anime-style faces. Dataset Preparation The dataset consists of preprocessed anime faces resized to 64×64 pixels. Each image is normalized to the range [-1, 1] using the formula: Images are loaded using ImageDataGenerator with the following setup: Model Architecture Generator The generator maps a 100-dimensional noise vector to a 64×64 RGB image using a series of transposed convolutions. Discriminator The discriminator uses Conv2D layers to downsample images and classify them as real or fake. GAN Training The discriminator and generator are compiled separately: Training Loop Results This project demonstrates how a DCGAN built with Keras and TensorFlow can effectively generate realistic anime-style faces from random noise. By leveraging transposed convolutions in the generator and convolutional layers in the discriminator, the model learns to produce increasingly detailed images over time. While basic in architecture, the results highlight the potential of GANs in creative AI applications. With further improvements such as advanced loss functions, deeper networks, and richer datasets, the quality and diversity of generated outputs can be significantly enhanced.
Brain tumors are abnormal growths of cells in the brain that can be life-threatening. Early and accurate detection is crucial for effective treatment. Deep learning, specifically convolutional neural networks (CNNs), has revolutionized medical imaging by providing automated and accurate diagnoses. This project focuses on detecting different types of brain tumors using a deep learning model trained on MRI images. Dataset The dataset used in this project consists of MRI scans categorized into three tumor types and one non-tumor class: The dataset is divided into: Data Preprocessing Model Architecture The deep learning model is built using VGG16, a pre-trained CNN model, with modifications for multi-class classification. The architecture includes: Training Process Results and Analysis Conclusion This project successfully demonstrates the potential of deep learning for medical diagnosis, particularly in multi-class brain tumor detection. The VGG16-based model effectively classifies MRI images into four categories with high accuracy. Future Enhancements: References:
With the advancement of technology, agriculture has seen significant improvements, especially with the integration of machine learning techniques. One of the pressing challenges faced by farmers is the early detection of plant diseases. This project focuses on building a Convolutional Neural Network (CNN) to classify plant diseases from images, specifically targeting diseases in corn, potato, and tomato plants. Dataset The dataset used in this project consists of images of plant leaves affected by three common diseases: The images were stored on Google Drive and loaded into the Colab environment for preprocessing and training. Data Preprocessing The preprocessing steps involved: Model Architecture The CNN model was built using Keras with the following structure: The model was compiled using the Adam optimizer with a learning rate of 0.0001 and categorical crossentropy as the loss function. Model Training The model was trained for 50 epochs with a batch size of 128. Training and validation accuracy were monitored throughout the process. The model achieved a satisfactory accuracy, as indicated by the plotted training history. Model Evaluation After training, the model was tested on unseen data to evaluate its performance. The results showed a high accuracy rate, indicating the model’s effectiveness in identifying plant diseases. Results and Analysis The classification report and confusion matrix revealed that the model performed well across all three classes. Additionally, the ROC AUC score demonstrated the robustness of the model. Conclusion This project successfully developed a CNN model to classify plant diseases with high accuracy. Early detection can help farmers take preventive measures, minimizing crop loss and ensuring better yield. Further improvements could involve using a more diverse dataset and fine-tuning hyperparameters for enhanced accuracy.