#pytorch

Simple, lightweight, and reproducible medical imaging AI toolkit.

Predictive GPU memory framework for AI inference workflows

Production-grade semantic caching for LLM applications using classifier-validated equivalence, cross-encoder verification and risk-aware cache policies.

A 1.32B-parameter Mixture-of-Experts LLM trained from scratch in PyTorch - custom tokenizer, MLA attention, 18B-token pretraining, SFT and DPO. No pretrained weights, no HuggingFace Trainer.

Manage AI workspace permissions on macOS. Control file access for Claude and Codex through a local, versioned runtime.

Top-down 2D 5v5 bomb-defuse multi-agent RL simulator with live match viewer. Recurrent MAPPO + TarMAC emergent communication, no scripted strategies.

Fine-tuning an open 7B LLM into a GST question-answering assistant, with a reproducible eval harness measuring the gain over the base model.

Production-ready Transformer implementation from scratch with multi-scale benchmarking and Arabic language support

A free mastery-based path from math and Python foundations to modern AI/ML engineering.

Faithful from-scratch PyTorch reproduction of OpenAI's GPT-OSS architecture (sliding/full attention alternation, learned attention sinks, YaRN 128K, top-2-of-8 MoE), scaled to Chinchilla-optimal 502M total / 247M active training on a single A100 80GB

A Privacy-Preserving Edge AI System for Trustworthy Safety Compliance

AI/ML & Software Engineering portfolio showcasing LLM systems, Agentic AI, RAG, ML infrastructure, distributed systems, and production AI projects

AI-powered music discovery platform with RAG pipeline, neural reranking, and real-time audio feature extraction. Describe any vibe, get a curated playlist.

Evidence-backed engineering portfolio focused on computer vision, deep learning, machine learning, computer networks, and digital and wireless communications.

SegAgent-AI: A Dockerized AI Platform for Industrial Defect Segmentation, Severity Analysis, and Cloud Deployment

Physics-informed AIML and geospatial decision-support system to identify urban heat hotspots, quantify microclimate heating drivers, and run optimized scenario-based cooling interventions for urban heat island mitigation.

AI-powered skin lesion classifier using ResNet18 transfer learning on HAM10000 dataset — PyTorch + Streamlit

A small GPT-style language model trained from scratch on TinyStories.

🌿 End-to-end full-stack AI Crop Disease Prediction & Agronomist Platform with 67 classes across 22 crops, Grad-CAM explainability, weather risk forecasting, and AI Chatbot.

101 Günlük Yapay Zeka, Bilgisayarlı Görü, LLM/RAG ve MLOps Mühendisliği Master Programı — MiniViT-MoE v2, FastAPI, Docker ve Uçtan Uca AI Sistemleri

A multi-agent LLM framework for explainable algorithmic trading on NSE-listed equities — DL forecaster, LLM news sentiment, and bull/bear debate, evaluated under a full Indian transaction-cost model.

Mac-local LLM-from-scratch lab for building and evaluating a small decoder-only language model on Apple Silicon.

AI-powered Voice & Text Chatbot built using Python, Streamlit, Hugging Face DialoGPT, and Deep Learning. Supports conversational memory, multiple chats, voice input, chat export, and real-time response generation.

CardioPulmoRAD-AI is a multimodal explainable AI framework for early cardiopulmonary risk stratification and tele-triage. It fuses chest radiographs (DenseNet-121) and 13 clinical vitals (MLP) with dual XAI (Grad-CAM + SHAP) to support frontline health workers and decision support including specialist-in-the-loop referral in low-resource settings..

Deep Q-Network (DQN) agent trained to play Flappy Bird using PyTorch & Gymnasium — with experience replay, target network, and epsilon-greedy exploration.

Production-grade distributed ML inference platform — FastAPI gateway, Redis-backed worker pool with dynamic batching, WebSocket result streaming, and a live ops dashboard. Serves YOLO, Whisper, RAG and text models, with an MCP agent server and streaming chat.

🐝 State-of-the-art Python AutoML metaframework — fuses GenAI (built on Firefly Agentic / Pydantic AI) with classical ML & Deep Learning. Hexagonal, secure-by-default, governed.

Reproducible QLoRA SFT and DPO experiments for open-weight LLM reasoning, with locked selection, matched controls, and audit-grade provenance.

Private, verifiable AI for business data. Verified answers copied from your records, deterministic HDC retrieval, exact aggregation, RBAC and audit trail. Local, offline, no cloud.

LocalSight · 198M-A64M 思考型 MoE LLM，2×RTX4090 从零训练（pretrain→SFT→SimPO→RLAIF→Agent RL），GGUF/Ollama 可运行

A 14M-parameter GPT built from scratch in PyTorch hand-written attention, custom BPE tokenizer, no pretrained weights.

From-scratch LLM inference engine — continuous batching, paged KV cache, prefix caching, hand-written Triton paged attention, CUDA graphs, OpenAI-compatible API. Benchmarked against vLLM on an A100.

Multi-camera person tracking & re-identification using YOLOv8 + OSNet, event-driven with Kafka, and streamed live via FastAPI WebSockets.

Camera-only congestion forecasting and RL signal control for unstructured Indian intersections. CNN-ViT cross-attention predicts per-lane congestion 60s ahead; a PPO agent uses that forecast to time the lights. A project - simulation-validated in SUMO.

Remove AI-generated content and chatbot widgets from websites with this open-source browser extension.

PPO agent trained from scratch to play a VizDoom deathmatch from raw pixels, with a convolutional actor-critic and shaped rewards.

A from-scratch PPO agent that acquires and holds a Pound-Drever-Hall laser lock, benchmarked against the sweep-and-capture servo every optical lab actually uses.

Readable architecture diagrams for PyTorch models. The tracer supplies the facts, an agent supplies the abstraction, and a coverage check proves nothing was silently dropped.

A MAPPO-based framework for intelligent, priority-aware routing in smart healthcare SDN networks.