Build AI at the
Speed of Light

25+ open-source projects across computer vision, language models, generative AI, and edge deployment — from research to production, all connected in one ecosystem.

Explore Projects Star on GitHub

                    
                    FlashDet
                    Real-time object detection, 298K→9M params

                    
                    FlashVLM
                    Vision-language model with DINOv2 backbone

                    
                    FlashStudio
                    Training & inference UI — pip install flashstudio

                    
                    FlashEdge
                    Edge deployment with TensorRT & ONNX

Ecosystem Clusters

Vision

6 nodes

Real-time object detection, segmentation, classification, OCR, pose estimation, and tracking.

FlashDet FlashSeg FlashCls FlashOCR FlashPose FlashTrack

GenAI

6 nodes

Large language models, vision-language models, diffusion, video generation, 3D, and audio.

FlashLLM FlashVLM FlashDiffusion FlashVideo Flash3D FlashAudio

Infra

5 nodes

Model optimization, edge deployment, inference engines, super-resolution, and model fusion.

FlashOptim FlashEdge FlashInfer FlashDownUP FlashFusion

Applied

5 nodes

Medical imaging, retrieval-augmented generation, autonomous agents, RL, and knowledge distillation.

FlashMed FlashRAG FlashAgent VisionRL DINO_Teacher

Tools

1 node

Interactive training studio with live metrics, model configuration, and one-click deployment.

FlashStudio

Code Flow

install

pip install flashdet

import

from flashdet import Predictor
detector = Predictor(
    pretrained_coco=True
)

predict

results = detector.predict(
    "image.jpg"
)
for cls, score, box in results:
    print(f"{cls}: {score:.2f}")

train

from flashdet import Trainer

trainer = Trainer(
    pretrained_coco=True,
    lora=True,
    epochs=50
)
trainer.train()

export

# CLI export
flashdet export \
    --format onnx \
    --half

# Launch Studio UI
pip install flashstudio
flashstudio

Performance Benchmarks

Speed vs Model Size

NVIDIA Jetson Nano — INT8 TensorRT

25020015010050

245 FPS

156 FPS

142 FPS

118 FPS

97 FPS

0.5M1.0M1.5M2.0M2.5M

Parameters

FlashCls-M 0.58M params

245FPS

FlashOCR-M 0.82M params

156FPS

FlashDet-M 0.95M params

142FPS

FlashSeg-M 1.1M params

118FPS

FlashDet-1.5x 2.44M params

97FPS

Pipeline Flow

Build 24+ architectures

Train LoRA + FlashStudio

Optimize FlashOptim

Deploy Edge + Cloud

Build AI at theSpeed of Light