Ashish Nitin Lodaya
AI Engineer & GenAI Strategist
I build intelligent systems that create real-world value. My work bridges the gap between advanced research and practical application, focusing on RAG, AI agents, and optimized models to solve meaningful problems.
About Me
My approach to building with AI.
From Idea to Impact
I'm an AI/ML Engineer who thrives on turning complex challenges into intelligent, user-centric solutions. With hands-on experience in both commercial and research settings, I specialize in designing and deploying systems—from autonomous AI agents to optimized Vision Transformers—that deliver measurable results, such as achieving 92% operational efficiency and 40% performance gains.
My core passion is creating meaningful value by applying AI thoughtfully and responsibly. Whether it's for education, finance, or productivity, I focus on building tools that are not only powerful but also practical and reliable. I'm always eager to collaborate, learn new techniques, and find innovative ways to solve problems with data and intelligence.
Technical Skills
A visual representation of my core competencies.
AI & GenAI Technologies
Retrieval-Augmented Generation (RAG), AI Agents, GPT-4, Claude, LangChain, LlamaIndex, Semantic Search
Machine Learning Frameworks
PyTorch, TensorFlow, Scikit-learn, HuggingFace Transformers
Programming & Deployment
Python, FastAPI, TypeScript, RESTful APIs, Docker, Microservices, CI/CD
Cloud & Databases
AWS (Bedrock, EC2, SageMaker), Vector Databases (FAISS, Pinecone)
Featured Projects & Research
A selection of work demonstrating my approach to building practical, high-impact AI solutions.
BIAgent: Business Intelligence Agent
An AI-powered platform that transforms natural language queries into actionable business insights. BIAgent features an advanced agentic layer where specialized AI agents collaborate to perform automated data analysis, predictive modeling, and interactive visualizations.
- Multi-Agent Collaboration: Specialized agents for data cleaning, pattern discovery, and visualization work in concert.
- Automated EDA & Modeling: Automatically performs exploratory data analysis and suggests predictive models.
- Natural Language Interface: Allows users to ask complex questions in plain English and receive interactive reports.
SmartCoach: Personalized Learning Agent
An intelligent AI agent designed to deliver personalized education. SmartCoach integrates real-time performance monitoring and goal tracking to curate and adapt learning content, ensuring an optimized and effective experience for each user.
- Adaptive Content: Uses reasoning to modify content based on user performance and feedback.
- Real-Time Monitoring: Tracks user progress and goals to dynamically adjust the learning path.
Architext AI: Diffusion Models for Design
Fine-tuned advanced AI models (SDXL) with LoRA adapters for controllable architectural generation, implementing a sophisticated prompt engineering pipeline using Flan-T5 and BERT to translate complex requirements into effective AI solutions.
- Controllable Generation: Leverages SDXL and LoRA for precise, style-controlled 3D design generation.
- Advanced Prompting: Integrates Flan-T5 and BERT to understand nuanced, detailed architectural descriptions.
- Real-Time Interaction: Deployed via FastAPI and Docker, enabling live interaction and generation on cloud infrastructure.
Research: ViT & Transformer Quantization
This research focuses on optimizing Vision Transformers for edge deployment. By applying an innovative Adaptive Logarithm Quantization (AdaLog) technique to a Dual-ViT model, we significantly reduced computational overhead without sacrificing performance on heritage site classification.
- 68% Model Size Reduction: Compressed the Dual-ViT model from 94.6MB to just 29.9MB.
- Maintained High Accuracy: Achieved 85.7% accuracy, nearly identical to the original 87.6% pre-quantization.
- Edge-Ready: Drastically lowered FLOPS, making the model suitable for resource-constrained devices.
Research: Hybrid Model Optimization
Developed a hybrid framework to optimize the RT-DETRv2 object detection model for edge devices. This approach strategically combines L1 unstructured pruning on the CNN backbone with 8-bit dynamic quantization on the Transformer decoder.
- Component-Specific Approach: Pruned the ResNet-18 backbone and quantized the decoder to maximize efficiency.
- 37% Size Reduction: Successfully compressed the model with only a minimal 5.5% drop in mAP.
- ONNX Deployment: Ensured seamless deployment and improved latency on edge hardware.
Project Impact Metrics
Career Journey
My professional experience in AI development and research.
AI Developer Intern
The Idea Company
- Achieved 92% operational efficiency via RAG & AI agents.
- Reduced model hallucinations by 30%.
Research Intern
KLE Tech University
- Achieved 40% performance improvement in model optimization.
- Designed robust object detection solutions.
Publications
My academic contributions and research publications.
IEEE Xplore, 2025
A comprehensive survey on optimization techniques for Vision Transformers in edge computing environments.
JEET Journal, 2024
This paper explores the intersection of human-centered design thinking methodologies and AI-driven automation.
Education & Certifications
Academic background and professional achievements.
Education
Bachelor of Engineering in Computer Science (AI)
KLE Technological University, Hubli
Aug 2022 - May 2026 | GPA: 8.77/10
Achievements
- DevOps Foundation Certified
- Smart India Hackathon 2023 - Top 15%
- Competed against 5000+ teams
Get In Touch
I'm always open to discussing new projects and opportunities.
Let's Connect
Have a project in mind or just want to say hello? I'd love to hear from you. Reach out via email or connect with me on my social platforms.