ADITYA KATKAR
Just Dreaming in Neural Nets
Building intelligent solutions that solve complex problems through code, creativity, and applied AI.
About
Call Me Addy :)
Portfolio
A curated collection of innovative problem-solving projects that reflect my perspective on design, strategy, and meaningful solutions.
Memory Pal-Your AI Brain for Everything You've Ever Read
MemoryPal is your AI brain — a context-aware agent that instantly becomes an expert on any topic or company, using the data you provide.Upload PDFs, emails, or web pages, and say: “Be my business consultant for this startup.”MemoryPal gets to work — pulling insights, tracking your questions, and citing sources.It turns static agents into smart, searchable collaborators with:.📚 Persistent memory across sessions.🔍 Timeline-based recall.📎 Source-linked answers.- Always learning.- Always growing.- Always ready to think with you
EduVerse-Your Personal Learning Universe, Anywhere
Adaptive Learning Engine. Dynamically adjusts difficulty based on student performance. Identifies knowledge gaps and creates personalized study paths Learns from student interaction patterns to improve over time.🌏 Cultural Context Awareness.- Automatically adapts content to local contexts (farming, markets, local currencies).- Uses familiar examples and culturally relevant scenarios.- Respects local educational traditions and knowledge systems.🎯 Socratic Teaching Method.- Guides students through discovery rather than giving direct answers.- Builds critical thinking skills through strategic questioning.- Encourages student confidence and independent problem-solving.📶 Offline-First Design.- Works completely without internet connection.- No data collection or privacy concerns.- Runs on modest hardware (laptop, tablet, or basic computer).🎨 Resource-Conscious Teaching.- Suggests activities using locally available materials.- Creates lesson plans for under-resourced classrooms.- Provides alternatives for expensive educational tools.
Indev.ai-Evolve and Evaluate the Startup
IndevAI bridges the gap between small startups, vendors, and investors by creating a single platform where innovation meets opportunity with AI-powered matchmaking:.🌟Startups get visibility, integrated AI assistance and AI-driven insights to present their business effectively.📊 Startups showcase verified pitch decks.🤝 Investors access structured startup profiles, market analysis, and risk assessments to make informed decisions using our custom Knowledge Base(which ranks out of 10).✨IndevAI simplifies startup discovery, fair access, faster discovery, and smarter investments, especially in emerging markets like India.
DeepRetrieval-An end-to-end model for high-precision, context-aware information synthesis.
DeepRetrival is a minimal, production-oriented implementation of a Retrieval-Augmented Generation (RAG) agent exposed through a Model Context Protocol (MCP) server, built without using any external systems.
IntelliLift-Control-System Using C++
A smart, efficient elevator management system that uses object-oriented design and algorithmic decision-making to control elevator movement, handle requests, log operations, and simulate real-time lift behavior.
BugHawk-Debug Smarter. Defend Deeper
We created BugHawk that helps debugging intensed based code quickly and also helps in maintaining that code with respect to future code frameworks and also helps in fighting against mischievous attacks on code.BugHawk is an AI-driven system designed to unify code debugging, maintenance, and security threat detection into a single intelligent pipeline.
Publications
A curated collection of my research interests and problem-solving projects that reflect my perspective on design.
Reconfigurable SoC Platform For Edge AI
- Publication: LAP LAMBERT Academic Publishing.- We developed low-power and reconfigurable System-on-Chip (SoC) on the Xilinx Zynq-7000 platform to provide real-time Indian Sign Language (ISL) recognition for the D/HH community.- Developed custom Verilog hardware accelerators using a pipelined dataflow architecture to deploy a quantized 8-bit LeNet-5 CNN, achieving a 99.54% recognition accuracy while operating entirely offline to eliminate cloud latency.
Transformer-Based Zero-Shot Learning for Robust and Scalable Product Classification
- Publication: ICEI 2026 (International Conference on Emerging trends and Innovations).- This was in track of Embedded Systems and Internet of Things.- Our goal was to demonstrate and implement transformer based model with zero shot prompt learning for classification based products that is practical approach to classify data that belongs to groups with no established learning background.
Aditya | AI & Full-Stack
Building optimized multi-agent systems and secure AI frameworks.
Scopus Publication
BugHawk: Automated vulnerability detection.
Scalable Transformers
ICEI 2026 Conference presentation.
Core CS & Optimization
Virtual memory simulations & C++ system architecture.
