ANALYTICS
Artificial Intelligence. For humans. By humans.
The bots are coming; well - they aren't just on the way; they are already among us. Leverage the power of machine-driven cognitive analysis to derive insights relevant to your industry. Be it image, video, audio, speech, text or an aggregration of data types, we have developed proven techniques that can extract actionable intelligence.
Beyond the hype - AI by Ignitarium
GENERATIVE AI
Innovate with Generative AI Magic!
Elevate your product engineering with our cutting-edge software services, powered by the transformative magic of Generative AI.
Our Expertise
- Use of Opensource models and mainstream GenAI models in API access mode
- Training LLMs with custom data (structured/unstructured)
- Finetuning and improving LLM performance
- Porting & optimization of LLM & LVM on AI HW accelerators
- Benchmarking various LLMs for performance
- Expertise in various Transformer architectures
- Prompt engineering
Applications
AI Model Benchmarking Framework:
Benchmark Categories
- Performance benchmarking
- Power benchmarking
- Various quantization levels
- Various batch sizes
- Various resolutions
Metrices used for Benchmarking
- Tokens-per-second
- Time-to-first-word
- Precision
- IoU, MAP
- Top1, Top5, Top10
Models used for Benchmarking
- LLaMA2
- MobileNet
- Resnet-50
- SSD
- Inception
- VGG-16
- AlexNet
- YOLO v2, v3
- Custom CNNs
- SqueezeNet
Model Frameworks
- Tensorflow, TF-Lite
- Caffe / Caffe2
- PyTorch, MXNet
- ARMNN, TensorRT
Platforms used for Benchmarking
- NVIDIA A100, H100
- NVIDIA RTX3xxx / TITAN
- NVIDIA Jetson Xavier / Nano
- Google Edge TPU
- NXP i.MX8M
- TI Jacinto7
- Intel Arria10 FPGA
- Custom AI accelerators
- Intel NCS
Data Management
Data quality is crucial in artificial intelligence because it directly impacts the performance, accuracy, and reliability of AI models.
Our team has expertise in
- Preprocessing data from various 2D and 3D sensors – camera, LiDAR and Radar
- Synthetic data generation
- Virtual sensors – using stored sensor data
- Automated data labeling
- Utilities for parameterized audio and video test stream generation
- Data versioning
Model Development
- Expertise in developing power and memory efficient Deep Learning (DL) based solutions using Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs), Transformers and Generative AI algorithms
- Expertise in developing classical machine learning algorithms such as Support Vector Machines (SVM), dimensionality reduction techniques, Markov models for solving classification, regression, and time-series forecasting problems
- We have capabilities to port AI models to all kind of edge devices and cloud platforms
- Model customization and graph surgeries to address layer support on different types of AI accelerators
- Model pruning and optimization to enhance the model runtime performance
AI SDK
We partner with AI chip companies who are in the process of developing AI hardware and software SDKs, and help reduce time-to-market for AI accelerators.
With our deep understanding of the customer’s hardware architecture, memory constraints, computational operations, our team can support:
- AI SDK kernel library development, enhancement, and porting
- Custom operator development and optimization for DL accelerators
- Full graph, sub graph porting and optimization
- DSP library porting and optimization
- Developing and optimizing OpenVX kernels, GStreamer plugins
Applications
- Architect, design and develop AI applications for various use cases across industries
- Application development on Edge devices and on cloud platforms
- With expertise in Embedded, AI, web development and cloud, we can develop complete solutions integrating device, cloud and user inputs
- Validation of algorithms, models and applications
Use Cases
Predictive Battery Analytics
Autonomous Mobility
Autonomous mobility is enabled by sensor fusion algorithms – which fuse multiple sensors (camera and LiDAR) using both classical & deep learning approaches, and navigation stacks.
Medical Imaging
The diagnostic accuracy and efficiency of medical imaging is being enhanced through advanced image analysis and pattern recognition algorithms, and also by automated patient positioning during scans.
Defect Detection
Ignitarium’s visual AI based defect detection platform, TYQ-i, captures visual data and uses AI models to detect defects in civil infrastructural assets like rail tracks and wind turbines.
WATCH VIDEO
Advanced Robotic Picking
WATCH VIDEO
Factory Worker Analytics
Factory Worker Analytics includes person and personnel protective equipment (PPE) detection such as helmet, jacket and goggles to optimize workforce productivity and safety. It is in-field trainable for new artifacts.
WATCH VIDEO
People Flow Tracking
People & object analytics in indoor and outdoor environments includes Counting and tracking, Flow monitoring, Dwell timing and Heat maps using 3D LiDAR and deep learning.
WATCH VIDEO
Surveillance
Surveillance in retail, employs aisle monitoring, shopper gaze detection, suspicious activity detection to analyze shopping patterns, optimize store layout and prevent theft.
AI : Platforms, Libraries & Frameworks
- AI FRAMEWORKS
- AI HARDWARE PLATFORMS
- CLOUD PLATFORMS
- LANGUAGES
- AI FRAMEWORKS
- AI HARDWARE PLATFORMS
- CLOUD PLATFORMS
- LANGUAGES