BreatheWish
AI-assisted chest radiograph screening and clinical triage platform featuring a DenseNet-121 classification backbone, Grad-CAM visual explainability, and a decoupled FastAPI clinical workflow service.
ML Engineer · Applied AI Systems
Building systems at the intersection of language models, dense vector retrieval, and production APIs.
I focus on moving from ambiguous real-world requirements to reliable production systems. My work spans data preparation, dense embedding retrieval, interpretability analysis, and decoupled API architectures.
Final-year B.Tech in CS (Machine Learning) at GPREC (CGPA 8.74 / 10), graduating 2027 · Datathon 2.0 winner at IIITDM Kurnool.
RAG library-policy chatbot featuring a FastAPI backend, FAISS + fastembed retrieval pipeline, and Vercel frontend. Grounded query-answering over official regulations.
AI-assisted chest radiograph screening and clinical triage platform featuring a DenseNet-121 classification backbone, Grad-CAM visual explainability, and a decoupled FastAPI clinical workflow service.
Self-evolving federated multimodal platform for decentralized clinical data processing and privacy-preserving model synchronization.
Interpretability research using the anthropics/jacobian-lens repository on Qwen models. Covers BBQ bias analysis, activation-steering experiments, and negative random-direction controls that overturned an early result.
Interactive library search and question-answering prototype built with Streamlit for V3 National Hackathon 2026; served as the origin prototype for Athenaeum.
Decision-support system combining productivity prediction, SKU clustering, and quantum-inspired order sequencing for warehouse fulfillment.
Sustainable e-commerce platform that surfaces environmentally responsible alternatives and transparent product criteria.
CNN feature extraction and FAISS vector index pipeline for visual nearest-neighbor image retrieval.
Built a content opportunity scoring pipeline using binary classification, applying precision@K as the evaluation metric; audited training data for label leakage and anomalies to safeguard prediction reliability.
Engineered BreatheWish, an AI-assisted chest radiograph screening and clinical triage platform featuring a DenseNet-121 classification backbone (90.9% test accuracy, 0.9906 ROC-AUC) with Grad-CAM visual explainability, FastAPI backend, and collaborative physician workflows.
Conducted interpretability research on the J-Lens Bias project using the anthropics/jacobian-lens codebase on Qwen models, running BBQ evaluations and negative random-direction controls.
Built a CNN embedding pipeline, FAISS nearest-neighbor indexing, and visual similarity retrieval workflow.
Developed rule-based conversational logic and TF-IDF / cosine-similarity content recommendation modules.
Engineered an autonomous hybrid Rules + ML revenue recovery agent using XGBoost and SHAP explainability, achieving an 86.4% recovery rate and 13.73% uplift over baseline.
Awarded first place for an ML-powered warehouse execution and sequencing system.
Built a RAG-powered library management prototype that evolved into Athenaeum.
Engineered scene-level semantic segmentation using SegFormer architectures.
I am interested in applied ML, retrieval-augmented generation, and backend systems collaboration. If you have an ambiguous problem or an engineering roadmap to discuss, reach out.
iamsyedhasnain04@gmail.com