At Avancio, we lead technological advancements, revolutionizing industries through expert-driven innovation. Our team thrives on solving complex challenges, making impactful changes across sectors like automotive, retail, and telecom.

We are always on the lookout for talented individuals with experience in the following areas:

ADAS/AD, Autonomous Navigation, Driver Monitoring Systems, or Vehicle-to-Everything (V2X) Communication.

Advanced Connectivity technologies, including 5G, Edge Networking, and IoT Connectivity.

AI/ML expertise in areas such as Computer Vision, Natural Language Processing (NLP), Predictive Analytics, and Reinforcement Learning.

Cloud and Edge Computing, with a focus on Cloud Migration, Cloud Security, or Platform as a Service (PaaS).

Cybersecurity, including Threat Detection and Response, Network Security, and Data Protection Compliance.

Future Mobility solutions, such as Connected Car Systems, Electric Vehicles (EV) technology, or Smart Traffic Management.

Experience in Industrializing Machine Learning, including Automated ML (AutoML), Model Deployment, and Scalable ML Pipelines.

Process Virtualization through Augmented Reality (AR), Digital Twins, and Workflow Automation.

Robotics and Industrial IoT, with a focus on Automation Solutions, IoT Device Management, and Predictive Maintenance.

Take on challenging and impactful projects that drive the future of tech.

Enjoy endless opportunities for learning and career progression.

Collaborate with a talented and passionate team of industry experts.

Benefit from a highly competitive salary and an exceptional benefits package.

Experience the freedom of flexible working hours within a supportive and dynamic work culture.

Access comprehensive health benefits and wellness programs designed with your well-being in mind.

Join us at Avancio to make a significant impact in the technology sector. If you're passionate about pushing boundaries and driving tech excellence, we invite you to become part of our dynamic team.

Open Application

The resource you're looking for isn't here. Let's get you back on track.

Page not found

In this article, we present how to get started with building a naive RAG with Llama-index and how to customize the retrieval process to obtain more relevant results. We finally elaborate on the importance of having an automated evaluation in place to choose which version should be deployed in production.

Retrieval-Augmented Generation (RAG) is a powerful framework used to enhance the contextual understanding of Large Language Models (LLMs) by incorporating additional knowledge. By extending the context provided to LLMs, RAG enables them to generate more accurate and relevant responses, making it particularly valuable for companies seeking personalized interactions with their data.

Numerous companies aim to utilize chatbots to interact with their clients. Consequently, the chatbot serves as the face of the company, embodying its capacity to offer assistance, convey warmth, and demonstrate dedication. Therefore, to ensure optimal customer experience and maintain strong client relationships, it is crucial to develop advanced chatbots that deliver exceptional performance.

The RAG process can be broken down into three main phases:

Indexing: External data is indexed by chunking the text from documents, computing embeddings, and storing them in a specialized database known as 

Retrieval: During retrieval, the embedding of a query is computed, and chunks that are most relevant to it are retrieved to serve as context for generating a response.

Generation: Leveraging the retrieved context, the LLM formulates its response.

Naive RAG implementation with Llama-index

To run the provided code, ensure you have Python 3.11.7 installed along with the following dependencies:

Download a sample text file of Paul Graham's essay for demonstration:

For this experiment, we will use OpenAI for the LLM and BGE as embedding model. 

 has been deprecated since the version v.0.10.0 of llama-index and replaced by 

 object enables to set global parameters and those are lazily instantiated. This means that only required elements will be loaded into memory.

With Llama-index, implementing a Naive RAG is incredibly simple:

While Naive RAG implementations offer a quick way to interact with documents, the quality of their responses is only as good as the context they retrieve. Tuning parameters like 

 can significantly impact performance. However, other approaches, such as Sentence Window Retrieval, have proven effective in retrieving the most relevant context.

 splits documents into chunks before computing their embeddings. However, you have the flexibility to control how your documents are parsed by selecting a different type of 

, which parses documents into single-sentence nodes while retaining the preceding and following sentences in the metadata field "window".

Setting a window size of 1 means that the "window" field will contain one sentence on either side of the original sentence. This approach significantly improves context relevance during retrieval, as it provides a more comprehensive context for each sentence.

 functions by swapping out the node's content with a specific field from the node metadata—referred to as "window" in this instance. 

In practical terms, while the retrieval process solely relies on the node content to locate similar nodes based on the query, it's the "window" metadata, comprising the original text and adjacent sentences, that's fed into the re-ranker. This input allows the re-ranker to evaluate the context and identify the nodes most valuable for addressing the query.

 class and leverages the BGE embeddings. This step selects the most informative nodes to answer the query, preventing the context window from being cluttered with irrelevant data. As a result, re-ranking often reduces the number of tokens used and decreases latency.

 most similar nodes to the query are initially retrieved. These nodes are then re-ranked based on their relevance to answering the query. Only the top 

 examples are retained and added to the context for generating the answer.

. This parser recursively split a document in chunks. This builds a hierarchy of nodes that are connected to their parent.

 recursively merges subsets of leaf nodes that are similar to the query. This helps retrieve as much informative context as possible.

Similar to the Sentence Window Retrieval approach, we implement a re-ranking step for the retrieved nodes to provide only the most valuable information to the LLM for answer generation. Both methods result in lower token usage and reduced response latency.

We conducted an experiment with three different RAGs and compared their results using the trulens-eval library. We devised 10 questions for each chatbot and evaluated the quality of their answers based on three criteria:

In this specific case, a very simplistic demo example, the primary improvement we observed was a significant reduction in token usage. This indicates that we achieved similar performances for a fraction of the initial cost. However, in larger examples, we can anticipate an improvement in Context Relevance as well.

In our upcoming episode, we'll delve further into our assessment procedure. We'll explore methods for automating it and determining which metrics are best suited for diagnosing our pipeline's issues, helping us identify areas that require more focus.

Llama-index Node PostProcessors documentation

Charles Camp is a seasoned ML engineer with nine years of experience, specializing in NLP. He's also an AWS Solution Architect Associate, skilled in deploying ML solutions.

After writing his Master thesis, using machine learning to build predictive models in the medical sector, at Carnegie Mellon University, Charles took on the challenge of combatting money laundering at Credit Suisse Zurich (Switzerland), where he first began utilizing NLP to identify negative news.

For the past four years, Charles has been freelancing, collaborating with startups and industry leaders to realize their NLP projects. During the last two years, he's been diving into the latest LLM tech, extracting valuable insights from company documents, building chatbots, and developing powerful recommender systems.

Explore advanced Retrieval-Augmented Generation (RAG) techniques using Llama-index. Learn how to implement naive RAG, enhance retrieval with sentence window and automerging methods, and evaluate performance. Discover strategies to improve context relevance, reduce token usage, and optimize chatbot responses for better customer interactions.

Boosting RAG Metrics: Two Practical Methods | Avancio

Boosting RAG Metrics: Two Practical Methods

In recent years, artificial intelligence and machine learning have made tremendous strides, revolutionizing industries from healthcare to autonomous vehicles. However, as these technologies become more sophisticated and ubiquitous, a critical question arises: How can we understand and trust the decisions made by complex AI systems?

This is where Explainable AI (XAI) comes into play. XAI refers to methods and techniques that make AI systems' decisions more transparent and interpretable to humans.  As AI systems become more complex and integrated into critical applications such as autonomous driving and healthcare, understanding the decisions made by these "black-box" models becomes essential. In this article, we'll explore the importance of XAI and delve into practical applications in two very different areas; traffic sign detection and lung cancer diagnosis to illustrate how XAI can be employed to enhance trust, improve models, and ensure safety and fairness.

The importance of XAI becomes evident when we consider real-world scenarios where AI decisions have significant consequences:

: In 2018, a Tesla operating on Autopilot was involved in a fatal crash in California. This incident highlighted the need for understanding how AI systems make decisions in critical situations. Understanding the ML models that power such systems can help in diagnosing and fixing issues, potentially preventing future accidents​.

: Large-scale biomedical studies, such as those conducted by Penn Medicine, are increasingly relying on AI for cancer diagnosis and treatment planning. The ability to explain AI-driven medical decisions is crucial for patient trust and regulatory compliance.

- By understanding how models arrive at their predictions, developers can identify and fix bugs, leading to more reliable AI systems.

- Transparent AI systems are more likely to be accepted and trusted by users and stakeholders.

- As AI becomes more regulated, the ability to explain model decisions will be essential for legal and ethical reasons.

Let's explore how XAI techniques can help us understand deep learning models used for traffic sign detection.

We utilize two Kaggle datasets for traffic sign detection: one with 57 classes and another with 14 classes. An existing model trained on the 57-class dataset achieves an impressive 0.99 F1 score. This model serves as the basis for our XAI exploration.

One way to understand model performance is through a confusion matrix:

Analyzing the confusion matrix helps identify which classes are often misclassified. This initial step is crucial in understanding model performance and areas that require improvement.

GradCAM (Gradient-weighted Class Activation Mapping) is a powerful tool to visualize which parts of an image influence the model's decision. By applying GradCAM to our traffic sign detection model, we can gain insights into what features the model focuses on when making predictions.

In this example, we can see that the model is focusing on the correct areas of the traffic sign to make its prediction.

Now, let's examine how XAI methods can provide insights into machine learning models for lung cancer detection.

We'll use a Kaggle dataset specifically designed for lung cancer detection. This is employed to train a model aimed at identifying cancerous patterns in medical images. Similar to the traffic sign detection case, we leverage XAI methods to interpret and validate the model's predictions.

Again, we start with a confusion matrix to understand overall model performance. The confusion matrix provides a detailed breakdown of the model's accuracy across different categories, highlighting its strengths and weaknesses:

SHAP (SHapley Additive exPlanations) is a game-theoretic approach to explain the output of ML models. SHAP force plots visually represent the contribution of each feature to a single prediction, while SHAP summary plots aggregate these contributions across the entire dataset. Let's look at some SHAP visualizations:

These plots show how each feature contributes to pushing the model output from the base (average) value to the final prediction.

Summary plots give an overview of feature importance and their impact on the model output.

This technique measures the importance of a feature by calculating the increase in the model's prediction error after permuting the feature's values:

These visualizations help identify which features are most crucial for the model's predictions, guiding further investigation and potential model improvements.

Explainable AI is not just a technical necessity; it's a bridge between complex AI systems and human understanding. By implementing XAI techniques such as GradCAM, SHAP, and permutation feature importance we can:

1. Improve model performance by identifying and fixing bugs.

2. Build trust in AI systems among users and stakeholders.

3. Meet regulatory requirements for AI transparency.

4. Gain valuable insights that can lead to scientific discoveries and better decision-making.

As AI continues to advance, the field of XAI will play an increasingly critical role in ensuring that these powerful technologies are used responsibly and effectively.

For those interested in diving deeper into XAI, here are some recommended resources:

Explainable AI: Interpreting, Explaining and Visualizing Deep Learning

The SHAP (SHapley Additive exPlanations) project

For a comprehensive view, check out the full project code and additional resources below.

As we advance AI capabilities, let's equally advance our ability to interpret and explain these powerful tools. It's not just about what AI can do, but how we understand its decisions.

Tesla in fatal California crash was on Autopilot: 

Discover the critical role of Explainable AI (XAI) in enhancing the transparency and trustworthiness of complex AI systems. Learn how XAI techniques like GradCAM and SHAP can improve model performance and compliance in critical applications, from autonomous driving to lung cancer detection. Explore practical case studies on traffic sign detection and lung cancer diagnosis, showcasing how XAI fosters safer, more reliable AI. Dive into the benefits of XAI, including building trust, ensuring safety, and meeting regulatory requirements.

Explainable AI: Unveiling the Black Box of Machine Learning Models | Avancio

Explainable AI: Unveiling the Black Box of Machine Learning Models

Small and Medium Enterprises (SMEs) across various sectors can greatly benefit from the Autonomous Mobile Robots (AMR). In warehouses and distribution centers, AMRs automate goods movement and order fulfillment, optimizing logistics with a goods-to-person approach. Manufacturing plants enhance their production workflow through efficient material transport and logistics. Small family farms and greenhouses utilize AMRs for moving crops, feed, and equipment, improving operational efficiency.

One of the most famous and successful stories of the early adoption of autonomous robots is the Kiva Robotics story. Founded in 2003 and acquired by Amazon in 2012 for nearly 800 million USD, they pioneered the industry and the goods-to-person concept.

Imagine transforming your business operations with a custom-built robot that you can assemble in just a few days. Picture the satisfaction of witnessing a robot you constructed autonomously navigating your facility, handling repetitive tasks, and allowing you to focus on more critical responsibilities. This isn’t merely a vision - it’s a project you can initiate today.

This series empowers you to build your own AMR using open-source designs and accessible manufacturing methods.

Follow our step-by-step guide featuring detailed drawings, 3D models, Bill of Materials (BOM), hardware architecture, navigation software and user interface packages. Utilize straightforward manufacturing technologies to integrate advanced automation seamlessly into your business operations.

The project aims to strike a balance between implementation complexity and achieving high-quality results. Our goal is to provide straightforward and cost-effective instructions for creating a professional-grade mobile robot suitable for real-world applications. Take it step by step and stay patient throughout the process.

Having mastered this guide, you can confidently advance this area independently and potentially enhance productivity not only within your enterprise but also for your partners.

In Picture 1, you can see what the completed project should look like. The conveyor is a separate feature and is not discussed in this article. However, you can independently install and connect any mechanism or extension to the mobile platform to give it the functionality you need. For this purpose, a mounting platform and holes for wire routing and connector installation are provided in the upper part of the robot.

If we look at other similar projects to create your own AMR, they often offer incomplete or limited solutions, for example, purely software options, devoid of electronics and mechanics, or kits suitable only for R&D and training due to the non-functionality of the body and mechanics. Commercial solutions often require additional payments for completion or use of the project.

However, it would still be beneficial to gain experience by researching projects like SEER, SlamTec, Roboteq, Ubiquity robotics, and Bosch Rexroth Rokit.

Our goal was to develop a simple, cost-effective and comprehensive solution for basic warehouse and production automation. Our design and software have successfully passed basic testing in real production conditions.

By open-sourcing our technology, we provide SMEs and individuals with an opportunity to leverage advanced robotics without the high costs associated with research and development.

Manufacturing, chassis and mechatronics: How-to guide for the OpenAMR robot

Perhaps one of the most important components of this series of articles is the design of the mobile platform. The project may seem cumbersome, and it is, because this project is closer to a professional one than an amateur one.

Don't be intimidated by the volume of information - you always need to eat an elephant in parts. Moreover, we will try to help you and answer your questions.

The key features and capabilities are described in Table 1.

Table 1 ⎼ Electromechanical characteristics

Picture 2 - Mobile robot assembly overview

Our design is optimized for manufacturability, requiring only basic technologies such as:

Using mostly 2mm thick metal sheets, the design is robust yet simple to produce, allowing one person to build the robot in just one day if every part is already produced.

The top of the robot features a mounting place for installing and connecting various external modules, such as:

Sheet metal parts vary by production. Currently, the parts are designed for the following coefficients and radii:

After painting, we install riveting nuts. It is allowed to install riveting nuts before painting, provided that the threaded part of the nuts is protected from coating.

Screws installed without lock washers - fix with a thread lock (eg Loxeal 55-03).

Production drawings, 3D models, STEP, and DXF files

Specification sheet, including all parts and assemblies, sensors, and fasteners, links to components

The metal frame of the robot (chassis) consists of 13 unique types of parts, totaling 20 individual components, including symmetrical and repeating ones. The chassis design and the recommended assembly sequence are illustrated in Picture 3.

Discuss the drawings from the folder “Production_files” with a contractor for cutting and bending metal. The document "

" and drawings together provides enough detail for any contractors to understand the requirements and produce it. Parts should be laser cut and bent as specified in the drawings.

Picture 3 - Mobile robot chassis overview

We aimed to minimize the number of parts, assemblies, and assembly operations as much as possible. However, there are still more than a dozen components. While we can't cover every assembly in this article, we will provide one example to guide you through the process.

In Picture 3, you can see the main parts and the chassis assembly. The number in the circle represents both the part number and the assembly step in the sequential assembly process.

We will examine the numbers marked in red in more detail in the following pictures. These steps represent the most challenging part of the assembly: assembling the drive wheels.

Detailed steps for assembling the drive wheels

Parts required: 1, 2, 3, 4, 5, BLDC motors, bearing housing sets, shafts, wheels, C-clip locking washers, fasteners.

Screw one of the bearing housing sets onto part 1.

Insert the shaft with a key through the two bearing housing sets and the wheel into the assembly (without fixing the second bearing housing set yet).

Secure the shaft by fixing it with a retaining C-clip washer (locking washer).

Use the BLDC motor attached to part 2 and insert the engine shaft with the key into the wheel axle.

Screw the BLDC motor and part 2 assembly to the part 1 and part 3 assembly.

This completed assembly can now be attached to base 4.

Follow the same steps to assemble the second wheel module.

After both wheel assemblies are installed on the base, proceed to the next step.

Install the wheel assembly on the central part 5. Fix everything together using screws to complete the chassis assembly.

The entire assembly process described above is also presented 

As already mentioned, to simplify this assembly to three steps, you can use a two-shaft motor from ZLTeсh, as shown in Picture 5.

This wheel motor has a power of 150 W and a diameter of 200 mm (8 inches). It includes a built-in 1024-bit encoder and comes with a driver from the same manufacturer. All documentation is available in the repository with datasheets in the 

. The motor wheel and its driver simplify the assembly of the robot and reduce the overall product cost. The wheels exhibit good wear resistance properties, and their characteristics match the specified claims.

Installation of rivet nuts, also known as threaded rivets, can be done using either a manual rivet gun or a pneumatic one.

If your bending machine has a bend radius different from the recommended one (depending on the tool and punch), you may need a file. However, to assemble a mobile robot, you will generally only need a set of screwdrivers, a rivet gun, and a set of wrenches. The operations are simple and intuitive.

From beginners to experts, don't miss this exciting step in your robotics journey.

Discover how SMEs across various sectors can boost efficiency with Autonomous Mobile Robots (AMRs). From automating logistics in warehouses to optimizing production in manufacturing plants and improving operations on small family farms, AMRs are transforming small businesses. Learn how to get started with this DIY guide for mobile robots.

Cookies