π B.Tech CSE (AIML) @ The Neotia University (CGPA: 9.42) Β |Β π West Bengal, India
πΌ AI/ML Research Intern @ NIT Durgapur Β |Β π¬ 4 Published Papers (IEEE & Springer)
Hey there! π I'm Arpan Pramanik, a passionate AI/ML Engineer & Full-Stack Developer focused on building production-grade SaaS AI applications that bridge cutting-edge artificial intelligence with scalable, enterprise-grade cloud architecture.
- π Production-Grade SaaS AI: Engineering end-to-end AI SaaS platforms featuring persistent vector stores (
pgvector/FAISS), multi-LLM failover routing, and real-time event streaming (SSE). - π§ Generative AI & Agentic Workflows: Architecting production RAG pipelines, autonomous AI agents, contextual Q&A engines, and document intelligence with LangChain, Groq & Ollama.
- π¬ Explainable AI (XAI) & Computer Vision: Developing custom deep learning architectures (multi-headed CNNs, EfficientNet + RBF-SVM) with Grad-CAM interpretability for real-world inspection.
- β‘ Scalable Full-Stack Engineering: Building high-performance, responsive web applications using Next.js, React, Node.js, Express & FastAPI with 3-tier authorization.
- π Enterprise Observability & Security: Implementing production-ready distributed monitoring, system tracing, and structured logging with OpenTelemetry, Prometheus & Pino.
- βοΈ Cloud & MLOps Infrastructure: Deploying scalable containerized & serverless microservices across AWS (EC2/S3/Lambda), Docker, Vercel & Railway.
Programming Languages
Frontend
Backend & APIs
AI, LLMs & Data Science
-00D9FF?style=for-the-badge){kind=icon}
Databases & Vector Stores
Cloud, DevOps & Observability
Tools & Platforms
|
|
π PaperLens AI β RAG-Powered Research Paper Assistant
AI-Driven Academic Research Platform with Persistent Vector Search & Multi-LLM Routing
- β‘ Production Vector Storage: Persistent vector search with Supabase
pgvector& FAISS with generator-based PyMuPDF parsing. - π― Citation Validation Pipeline: 4-stage automated validation (
DOI β Exact β Title β Loose) via Semantic Scholar API with real-time Server-Sent Events (SSE) progress streaming. - π§ Resilient Multi-LLM Routing: Automatic failover powering 6+ AI research modules (Contextual Q&A, Experiment Planner, Research Gap Detection, Problem Generator).
π FruitQ-GradeX β Explainable AI Quality Assessment SaaS
IEEE Published Multi-Head Computer Vision Framework with Grad-CAM Explainability
- ποΈ Multi-Head Deep Learning: Custom CNN architecture achieving 97%+ test accuracy for joint fruit classification & quality grading.
- π¬ Grad-CAM Explainability: Integrated visual attention heatmaps visualizing model decision rationale, deployed via Streamlit.
π SSH β Observable Student Management Platform
Enterprise Multi-Portal SaaS with Full Observability Telemetry
- π 3-Portal Microservices: Admin, Faculty, and Student portals with 25+ REST API endpoints engineered for NAAC/AICTE compliance.
- π‘οΈ Production Observability: Integrated JWT auth, strict CORS, rate-limiting, Pino structured logging, and OpenTelemetry continuous tracing.
- π₯ Authors: Shibdas Dutta, Subhrendu Guha Neogi, Diya Chanda, Arpan Pramanik, ΓzgΓΌn Girgin, Enes Ladin ΓncΓΌl
- π DOI: 10.1109/ICRITO66076.2025.11241706
- π‘ Summary: Multi-task deep learning framework for fruit classification and quality assessment using multi-headed CNN. 98% classification & 99% quality detection accuracy with Grad-CAM interpretability.
π Paper 2: CropSense: Explainable Deep Learning Framework for Accurate Quality Detection in Solanaceous Crops
- π₯ Authors: Shibdas Dutta, Subhrendu Guha Neogi, Shiladitya Chowdhury, Vikrant Chole, Arpan Pramanik, Diya Chanda
- π DOI: 10.1109/ICRITO66076.2025.11241535
- π‘ Summary: Lightweight multi-headed CNN for potato and tomato quality classification. 99.9% crop classification & 98.5% quality detection accuracy with Grad-CAM, deployed via Streamlit.
- π₯ Authors: Shibdas Dutta, Barshan Adhikari, Arpan Pramanik, Diya Chanda
- π DOI: 10.1109/COMPUTINGCON64838.2025.11376762
- π‘ Summary: Hybrid CNN-ViT model for simultaneous crop classification and quality assessment. Reduces parameters by 30%+. 98.45% potato & 97.49% tomato classification accuracy with 98.5% quality assessment.
π Paper 4: Hyperspectral Fruit and Vegetable Classification Using Convolutional Neural Networks with EfficientNetB3
- π₯ Authors: Shibdas Dutta, Subhrendu Guha Neogi, Arpan Pramanik, Diya Chanda, ΓzgΓΌn Girgin, Enes Ladin ΓncΓΌl
- π DOI: 10.1007/978-3-032-21901-5_35
- π‘ Summary: Transfer learning framework with EfficientNetB3 across 4,320 hyperspectral images over 36 classes. 99.29% training & 97.21% test accuracy for intelligent sorting, deployed via Streamlit & OpenCV.
| Domain | Skills |
|---|---|
| Machine Learning | Supervised/Unsupervised Learning, Feature Engineering, Model Optimization |
| Deep Learning & XAI | CNNs, EfficientNet, Transfer Learning, Grad-CAM (XAI), Image Classification, Computer Vision |
| LLMs & AI Agents | LangChain, Groq, Ollama, RAG Pipelines, FAISS, Vector Search, Prompt Engineering |
| Web Development | MERN & Next.js Stack, RESTful APIs, SSE, Authentication (Clerk, JWT), Admin Dashboards |
| Database & Vector DBs | Supabase (pgvector), PostgreSQL, MongoDB, Redis, Pinecone, ChromaDB, MySQL |
| Cloud, DevOps & Observability | AWS (EC2/S3/Lambda), Docker, GitHub Actions, OpenTelemetry, Prometheus, Pino Logging |
| Data Science | EDA, Data Visualization, Statistical Analysis, Pandas, NumPy, SciPy |
πΌ Open to: Internships, Collaborations, Freelance Projects
β‘ "Code is like humor. When you have to explain it, it's bad." β Cory House