About
I build AI systems that work — not just in notebooks, but in production pipelines, real hardware, and agentic workflows that replace manual effort with intelligent automation. I'm a final-year B.Tech (Electronics & Computer Science) student at RCOEM, Nagpur, with hands-on experience across the full AI stack: from fine-tuning and prompting LLMs (Llama 3.3 70B, Qwen 2.5 Coder) on Groq, to deploying multi-agent code explainer and README generator systems, to integrating AI with ESP32-based IoT sensors on FastAPI backends. My research work includes a published-track paper introducing a Hybrid Stylometric-BERT model for AI-generated text detection — achieving 99.44% accuracy on ~114K samples by combining 50 stylometric features with frozen BERT embeddings via an MLP classifier, targeting Springer's IJIT journal. Currently, I'm developing an AI-powered industrial IoT monitoring platform targeting pharma cold storage — using AI agents as autonomous "employees" to monitor, alert, and act on sensor data with minimal human intervention. What I bring: → LLM & GenAI: Prompt engineering, RAG pipelines, multi-agent orchestration, Groq API → AI/ML: Python, Scikit-learn, BERT, Pandas, NumPy, Jupyter → Backend + IoT: FastAPI, ESP32, DHT22 sensors, REST APIs, React dashboards → Research mindset: interpretable ML, hybrid architectures, real-world validation I'm actively seeking Trainee AI Engineer, GenAI Engineer, and ML Engineer roles — remote or India-based — where I can build real AI systems that create measurable impact. 📩 vinitkhapekar09@gmail.com · Open to opportunities
Work experience
Research Intern at Shri Ramdeobaba College of Engineering and Management, Jan 2026 - Present · 5 mos, • Curated and merged 3 public datasets into a 114K+ sample dataset; prevented data leakage using prompt-based stratification • Developed a TF-IDF + Logistic Regression baseline achieving 94.04% accuracy • Fine-tuned a BERT-based classifier achieving 98.77% accuracy, 0.9868 F1-score, 0.9995 ROC-AUC — a +4.73% improvement over baseline • Evaluated models across accuracy, precision, recall, F1-score, and ROC-AUC; conducted full comparative analysis
