Shubham Kukadiya
M.Tech–MachineLearning
Experience
Oracle Consultant-Human Capital Management and Agentic AI
Mastek
Ahmedabad
Implemented a full enterprise structure for a multi-subsidiary holding company and engineered a production-style AI Agent with a 10-step orchestration flow for complex HR workflows like automated resignations. Architected a multi-tool LLM agent integrating 13 heterogeneous APIs; reduced input token count by 88% (from 42,000 to under 5,000) via a single-call startup strategy, eliminating context window crashes. Designed and enforced rigorous safety guardrails (ID masking, out-of-scope refusal, and conflict-gated flow) validated through exhaustive edge-case testing across 10 distinct employee personas.
Junior Research Fellow-Deep Learning for Precise-Direct Positioning in 6G Networks
Smart Energy Learning Centre (BSES)
Gandhinagar
Engineered a custom Transformer-based deep learning model for precise indoor positioning by tokenizing multi-channel wireless signal data (Channel Impulse Response and Reference Signal Received Power) for robust feature extraction. Developed an encoder-only Transformer architecture with a specialized Multi-Layer Perceptron (MLP) output head for direct regression, mapping intricate wireless channel data to precise (x,y) user coordinates.
Frontend Developer-React.js(MERN)
AppOpener
Hybrid(Jaipur)
Revamped the company’s landing page using React.js, creating a modern, responsive UI that significantly boosted user engagement and session duration. Optimized real-time responsiveness and reduced link generation latency by over 30% by integrating performance-focused React libraries. Refactored and debugged backend routing logic, resulting in faster redirection times and a seamless end-to-end user experience across devices.
Education
Dhirubhai Ambani University
M.Tech
CPI:8.47
LNMIIT
CPI:6.35
12th School SR Public School (CBSE)
Percentage:85.6%
Silver Bells Public School (CBSE)
Percentage:95%
Skills
Publication
A Hybrid CNN-Transformer Framework for Precise Indoor Positioning in 6G Networks
July2026
FirstAuthor|IEEEInternationalMediterraneanConferenceonCommunicationsandNetworking(MeditCom),Cagliari, Italy •Architecture Design:EngineeredFusionLoc, a hybrid framework synergizing 1D-CNNs with a geometry-aware Transformer encoder to fuse multi-modal 6G signals (CIR and RSRP). •Model Innovation:Introduced a spatial encoding mechanism to inject 3D TRP coordinates into attention layers and a custom “tail-aware” loss function to optimize for worst-case (P 90 ) reliability. •Results:Achieved decimeter-level precision (0.085m RMSE, 0.129mP 90 ) on the 3GPP InF-DH benchmark, surpassing traditional geometric and baseline deep learning methods.
Projects
YouTube Transcript RAG|LangChain,FAISS,HuggingFace,Groq/Qwen
Developed an AI chatbot that answers questions from YouTube videos using their transcripts by combining retrieval-augmented generation (RAG)withLangChainandFAISS. Implemented an end-to-end pipeline tofetch transcripts,split text,generate embeddings(via sentence-transformers/all-MiniLM-L6-v2), andstore vectorsin FAISS for fast semantic retrieval. IntegratedChatGroq (Qwen-3-32B)as the LLM to provide context-aware, transcript-grounded answers with concise and accurate summaries.
Agricultural Surveillance System|Python,PyTorch,UltralyticsYOLO,SwinTransformer,OpenCV
Architected anend-to-end UAV image-analysis pipelinefor real-time crop-health assessment and weed detection usingYOLOv8mandOpenCV. Integrated aSwin Transformer backboneinto YOLOv8m and applied a curriculum-style training schedule (head-only warm-up→full fine-tuning), increasingmAP50 from 0.875 to 0.905. Validatedcross-dataset robustnesson an external weed corpus (mAP50 = 0.889) and optimized model inference to 20 ms/imagefor near-real-time operation.
Automated Job Scraper|n8n,GoogleGemini,RapidAPI,GoogleSheets
Developed ann8n workflowto automate job scrapingfrom RapidAPI, implementingAI-powered summarizationof job descriptions using theGoogle Gemini chat model, and configureddata processing pipelinesto save structured job details and summaries toGoogle Sheets.
AI-Driven Book Recommendation System|Python,Pandas,Scikit-Learn,Flask,React(TypeScript),TailwindCSS
Processedlarge-scale book datasetswithPandasto extractuser preferences and rating patterns, enabling personalized recommendations. Engineeredcollaborative filteringandcosine similarityalgorithms inScikit-Learn, exposing results through aFlask REST API. Developed aresponsive React (TypeScript) frontendconsuming the API, with amodern UIbuilt usingTailwind CSS.
Certifications
Udemy Certification (2025):Complete Guide to Building, Deploying, and Optimizing Generative AI with Langchain and Huggingface
Udemy Certification (2024):Complete A.I. & Machine Learning, Data Science Bootcamp
Udemy Certification (2023):The Complete Full-Stack Web Development Bootcamp
Position Of Responsibility
Member, Literary Club
2020 – 2022
LNMIITStudentClub • Organized and facilitated literary activities including book discussions and reading circles to foster critical thinking and engagement among students. • Collaborated with peers to host cross-disciplinary events promoting literature within scientific communities.