AI/ML Engineer in the Making | I Build Models Where the Cost of Being Wrong Matters | Python . DSA . Machine Learning . NLP . RAG & LLM model . AI Agents
I'm Harsh a Pre final-year AI/ML student with a 8.8 CGPA and a deep obsession with understanding why a model does what it does, not just whether it performs well on a test set. My interest sits at the intersection of Algorithms and Statistics ,NLP , Transformers , RAG System and Responsible AI fields where the stakes are high enough that if it doesn't works isn't a good enough answer. I write clean Python, think in systems, and I'm genuinely comfortable going from raw data to a deployed, explainable output. But more than the technical stack, what drives me is building things that are honest about what they can and can't do, models that a domain expert can question and a non-technical stakeholder can actually understand. I'm currently looking for Data Science or ML roles where I can bring that mindset into a real team, on real problems If you're building something that needs that kind of thinking, let's talk.
Zidio Development
Aug 2025 - Oct 2025 · Bengaluru
1) Built an end-to-end image captioning pipeline using a CNN-LSTM encoder-decoder architecture, trained on the MS COCO dataset (83K+ images) with PyTorch 2) Engineered a transfer learning strategy by fine-tuning a pretrained ResNet-50 encoder while training the LSTM decoder from scratch, reducing training time significantly compared to full model training 3) Diagnosed and resolved an overfitting issue mid-training by iteratively tuning embedding size from 512 → 256, recovering coherent caption generation on unseen test images 4) Deployed an interactive captioning demo using Gradio, enabling real-time image-to-caption inference accessible via local web interface
Zidio Development
08/2025 – 10/2025
Built an end-to-end image captioning pipeline using a CNN-LSTM encoder-decoder architecture trained on MS COCO with 83K+ images, delivering real-time inference via a Gradio web interface. Reduced training time through a transfer learning strategy using a pretrained ResNet-50 encoder while training the LSTM decoder from scratch. Collaborated with cross-functional teams to define pipeline requirements and applied Agile iteration cycles to deliver a demo-ready captioning system within the internship timeline.
Bachelor of Technology - BTech, Artificial Intelligence and machine learning
Aug 2023 - Jun 2027
B.Tech, AI & ML Engineering
2023 – 2027
CGPA: 8.77
Built an end-to-end Retrieval-Augmented Generation chatbot using generative AI models and a Pinecone vector store to deliver context-aware responses on medicines and drug interactions. Designed prompt engineering strategies to optimize LLM output quality, grounding responses in a structured medical encyclopedia knowledge base.
Trained a CNN-based deep learning classifier on brain MRI scans across four categories—glioma, meningioma, pituitary, and no tumor—achieving 95%+ test accuracy via transfer learning on VGG16.
Architected a 4-stage agentic workflow—Search, Reader, Writer, and Critic—using LangChain tool-based agents, enabling automated information retrieval, synthesis, and evaluation at scale.