An Indian startup builds affordable edge AI boards for robots, drones, and industry, positioning them as Raspberry Pi alternatives. In an EFY interview with Nidhi Agarwal, Vicharak’s Akshar Vastarpara discussed the edge AI SBCs and innovations in physical AI systems.
Q. What does your company do?
A. At Vicharak, we build edge AI computing hardware and software platforms for the next generation of physical AI applications such as robotics, drones, autonomous systems, and smart industrial solutions. We develop our own AI-enabled single-board computers, similar to Raspberry Pi or NVIDIA Jetson platforms, but with built-in AI acceleration and a strong software ecosystem and community support. Our primary focus is edge computing, which means AI processing happens directly on the device or on-premises instead of relying on cloud platforms like AWS or Google Cloud. Our platforms are optimised for efficient AI workloads using CNNs, lightweight LLMs, and other neural network models, enabling low-latency, privacy-focused, and scalable AI deployment at the edge.
Q. Can you tell us about your product line?
A. We started with FPGA-enabled solutions because FPGAs play a critical role in chip prototyping; they allow designers to validate, emulate, and test chip designs before manufacturing. Our goal was to make this technology accessible not just to engineers, but also to students, hobbyists, and early-stage innovators. The biggest barriers in this space were cost, complex software tooling, and the difficulty of programming. To address this, we built Shrike-lite, an Arduino-plus-FPGA board priced at just ₹369, making it one of the world’s most affordable FPGA boards. It is designed for beginners, from school students to engineering learners, who want to explore electronics, computing, and chip design easily and practically. The response has been strong, with nearly 5000 units sold across India and international markets like Japan, the US, Europe, and Africa.
Another major product in our lineup is Vaaman, which combines a single-board computer with an FPGA in the same form factor, offering users much higher-performance computing possibilities compared to traditional solutions like Raspberry Pi. Beyond these, our product ecosystem is organised into three main series: Vaaman, Shrike, and Axon, each designed to serve different user segments and use cases. These are just the starting points, we are already expanding with products like Vaaman 2 and upcoming Axon variants such as Axon Lite, Axon Mini, and Axon Pro. Alongside hardware, we are also building software support, example solutions, and scaling manufacturing and marketing to strengthen the ecosystem.
Q. How do your SBCs support the physical AI revolution?
A. Physical AI is about creating systems that can sense their environment, process data, make decisions, and act in the real world, such as factory automation robots or smart robotics applications. To build these systems, developers need a Linux-enabled computing platform capable of running Python scripts or frameworks like Robot Operating System, along with built-in AI capabilities for vision, image, audio, and sensor-data processing. Our SBCs are designed as complete edge-computing platforms that combine processing power, memory, storage, connectivity, and AI acceleration in a compact form factor, making it easy to start building and deploying physical AI applications.
What makes our platform unique is its ability to handle the full physical AI pipeline on a single board, collecting data from cameras, microphones, and sensors, processing it in real time, making intelligent decisions, and then driving motors or actuators through GPIO and signal interfaces. Our SBCs are specifically optimised for robotics, automation, and edge AI use cases, enabling developers to build intelligent systems faster and with fewer hardware dependencies.
Q. How are your SBCs different from Raspberry Pi, and why should people buy them when Raspberry Pi is already so widely available?
A. Raspberry Pi was originally built mainly for hobbyists, education, and lightweight edge computing, which makes it a strong choice for simple Linux or Python-based applications, basic automation, or single-camera low-resolution use cases. But as edge applications become more advanced, especially with AI, computer vision, and industrial workloads, their limitations become clear. Raspberry Pi does not have built-in AI acceleration, so users often need extra hardware like AI HATs, and its computing and I/O capabilities are comparatively limited. Our SBCs are designed to address this gap with higher processing power, dedicated AI acceleration built in, stronger multi-camera support, expanded storage options, richer display connectivity, and wider sensor integration.
The reason customers choose our SBCs is flexibility across different levels of complexity and budgets. If someone needs simple, low-cost computing, Raspberry Pi can still be a good fit. But for users building production-grade edge solutions that need more performance and native AI capability, our boards offer significantly better value. We are also launching Axon Lite, which matches Raspberry Pi’s price segment while adding built-in AI capabilities, making it a superior option at the same budget. On top of that, our upcoming modular SBC platform allows users to customise connectors and hardware modules based on their application, something very different from Raspberry Pi’s fixed design. With products ranging from ₹7000 to ₹40,000, we cover everything from budget-friendly AI devices to high-performance industrial solutions.
Q. How does a system with both a CPU and an FPGA on the same board divide tasks, especially for AI workloads?
A. In a CPU+FPGA system, the CPU usually handles general-purpose tasks such as system control, software logic, and sequential processing, while the FPGA takes on tasks that benefit from parallel processing and low-latency execution. For AI workloads, the FPGA can accelerate inference and specific AI pipelines more efficiently than a CPU, though GPUs are generally stronger for large-scale AI computation. The key advantage of an FPGA is that it can be reconfigured at the hardware level to optimise data flow and computation for a specific task.
FPGAs are not limited to AI, they are especially useful in applications like radar, telecommunications, and industrial automation where ultra-low latency and parallel execution are critical. For example, in machine automation systems running at 300–400 FPS for simple classification tasks such as sorting objects by colour, FPGAs can outperform GPUs because they provide faster response times and direct control over hardware. The choice between CPU-only, CPU+FPGA, or other platforms ultimately depends on the performance, latency, and flexibility requirements of the application.
Q. What programming languages are used for FPGA programming?
A. Unlike microcontrollers, which can be programmed in languages like Python, C, or C++, FPGA programming is mainly done using hardware description languages such as Verilog, SystemVerilog, and VHDL. Instead of writing software instructions, these languages describe the digital circuit itself at the RTL (register transfer level), which is then converted into a bitstream that the FPGA can understand.
The real challenge in FPGA development is not learning the language but managing the toolchain. Traditional FPGA vendors like Xilinx and Intel provide powerful but very large software suites, which can be heavy for students and beginners. Modern FPGA platforms simplify this by offering lightweight tools and cloud-based synthesis, allowing users to write HDL code in a web browser and compile it online without depending on bulky vendor software or specific operating systems.
Q. Can computer science engineers or other non-hardware users also use this FPGA CPU board, and is it accessible for beginners?
A. Yes, absolutely. Our goal is to make the board accessible not just for hardware engineers, but also for computer science engineers and other non-hardware users. We have made the examples much easier to understand and use. Everything is open source on GitHub, and we currently have around 25 examples available, each with clear descriptions and documentation. This makes it possible for even non-electronics or non-engineering users to get started and learn easily.
Q. Are all your boards and software open source?
A. Not all of them. Right now, the Shrike-lite is open source, and most of the software is also open source. However, the VMware Axon1 Series hardware itself is closed source, while the Linux layer remains open source. The long-term goal is to keep the software ecosystem open, especially around FPGA tools and IP. In an industry where even basic IP often comes at a cost, the IP developed for these boards is provided free forever, so users don’t have to pay extra software licensing fees on top of the hardware.
Q. Who are your primary and targeted customers, and which focus areas do you serve?
A. We don’t have a single fixed customer type, we serve a wide range of users. This includes people working in robotics, such as humanoid robots, machine intelligence systems, and factory automation. We also support users focused on machine vision applications like human counting, face recognition, posture detection, and industrial inspection systems. In addition, students and hobbyists developing robotics or AI-based projects use our platform, as well as those working on autonomous vehicles.
Q. What microcontroller chips and FPGA families are you using in your boards?
A. We use different processors and controllers across our designs: in the Shrike series we use the RP2040, a Cortex-M based microcontroller, along with an FPGA from Venus, and we are also planning to launch a Wi-Fi enabled FPGA board combining ESP32 with FPGA by the end of this month; in Shrike-Fi and Vaaman we use the RK3399, a TSMC-fabricated 6-core processor paired with an FPGA from a US-based company focused on low-end FPGA solutions for edge computing; additionally, we are moving toward Qualcomm-based single board computers and plan to launch the Axon Mini based on the Qualcomm 366490 platform in the coming month or next month.
Q. What design challenges did you face while developing these boards, and how did you resolve them?
A. The main design challenges came from increasing complexity as we moved from simpler 10-layer boards like Vaaman to 12-layer systems like Axon, and eventually to advanced HDI PCBs with 14–18 layers. Initially, we didn’t fully understand what we were designing and lacked experience with high-speed design practices. We had to learn things like RAM-to-CPU routing, power integrity, high-speed signal simulations, and constraints like blind and buried vias. A major challenge was also that manufacturers would often flag limitations during design reviews, forcing us to revise layouts. Additionally, even small changes could take 5–8 days due to repeated validation and fabrication cycles.
We resolved these challenges mainly through continuous learning and iteration. We started using online resources, courses, and documentation to understand simulation techniques and design rules. Over time, we adopted RAM and power simulations to validate designs before fabrication, which reduced failures. We also gained experience in board bring-up, where hardware and software integration issues are identified and fixed after testing. As a result, our iteration cycle improved significantly, what initially took 1.5 years for Axon was reduced to around 4 months for Axon Lite because we had built a better understanding and faster validation workflows.
Q. What design and integration challenges did you face while building a product that combines hardware, software, and AI, and how did you handle them?
A. The main challenge in integrating hardware, software, and AI is that each team often assumes the issue lies on the other side, hardware teams believe their system is correct, while software teams believe their code is correct. In reality, problems can come from either side or from their interaction, which makes debugging difficult. Unlike pure software systems, hardware–software integration issues are harder to trace because you can’t easily inspect or probe everything, and resolving some issues can take days or even weeks due to the complexity involved.
Another major constraint is that hardware fixes are slow and expensive because every change requires a new iteration, which impacts both time and cost. Because of this, the focus is usually on solving as much as possible through software workarounds or adjustments first. However, some problems inevitably require hardware modifications. The key is tight collaboration between teams, continuous communication, and jointly figuring out whether the root cause is in hardware, software, or both.
Q. What testing approach are you using for your hardware and PCB manufacturing, and do you use simulation on the software side?
A. We currently use a mostly manual testing approach for our PCB and hardware systems. Since industrial-grade automated test chip setups are very expensive (around ₹20–30 million), we instead built our own software-driven manual testing framework. In this setup, each peripheral, such as camera, audio, storage, Wi-Fi, Bluetooth, speaker, and microphone, is tested individually using scripts and connected hardware to verify proper functionality. Although this approach is time-consuming, our strong in-house hardware team has also developed low-cost custom testing solutions to improve efficiency, allowing us to test multiple units quickly.
For example, we have created a dedicated testing system for products like Shrike-lite that can test 4–8 units per minute, enabling up to ~500 units per day per person. Alongside physical testing, we also use simulation tools on the software side before manufacturing, including power analysis, RAM analysis, and other pre-hardware validation techniques. As we scale, the plan is to gradually move toward more fully automated testing systems.
Q. Are there any current limitations you are addressing or improvements you are making in your upcoming boards?
A. Yes, we are addressing several current limitations in our upcoming boards, mainly focusing on three areas: first is hardware rigidity, where earlier users were stuck with fixed connectors like Ethernet or USB and had to replace the whole board for upgrades, so we are introducing a modular approach that allows custom connectors and peripherals; second is power efficiency, especially for AI workloads, where we are developing low-power solutions (including upcoming Qualcomm-based platforms) that can run efficiently even within 5V or similar constraints; and third is software experience, where unlike many complex boards that lack strong software support beyond Raspberry Pi, we are building a much more integrated ecosystem with features like remote access, cloud platforms, browser-based control, screen sharing, and more, aiming to combine flexible hardware with a strong, user-friendly software stack similar to an Apple-like approach.
Q. What inspired you to start the company, and what is the story behind its name?
A. We started the startup out of a deep interest in computation, understanding how CPUs, GPUs, and computing systems work at their core. While exploring this space, we saw an opportunity with FPGAs and emerging hardware platforms to build solutions around fundamental computation. That’s also why we chose the name Vicharak, which means ‘someone who thinks,’ because computation, at its essence, is about thinking and processing information, which aligns with our vision.
Our journey has been very challenging since we came from a software background and moved into hardware. Hardware required upfront investment, multiple iterations, and learning everything from PCB design to system bring-up. We faced manufacturing failures, funding constraints, and early scepticism about Indian hardware, but we kept pushing through each problem. Over time, by focusing on engineering, improving local manufacturing, and learning from failures, we built capability in India and gradually gained acceptance in both domestic and international markets.
Q. Can you explain your current manufacturing setup and how you handle production, testing, and future scale-up?
A. Right now, our primary focus has been on completing the product and bringing it to market. At present, we use a contract manufacturing facility in Baroda, near Surat, for the SMT pick-and-place process. We do not yet have our own pick-and-place line, so this part of the production is outsourced. However, we design the complete board, source all the components ourselves, and handle the majority of the production workflow in-house.
The outsourced assembly process accounts for only about 30% of the total manufacturing, while the remaining 70%, including debugging, testing, firmware loading, packaging, and quality control, is done by our team in Surat. If there are any manufacturing defects, we handle the fixes ourselves since the contract manufacturer only performs assembly. Going forward, we plan to set up our own pick-and-place line within the next 3–4 months, starting with the Shrike series and gradually expanding to products like Axon, Vaaman, and other complex products.
Q. Have you faced any supply chain issues in your hardware manufacturing process, and how did you handle them?
A. Yes, we did face a major supply chain issue when we moved from low-scale production to mass manufacturing. Around that time, a global DRAM shortage occurred due to the AI boom, which caused a sharp rise in memory chip prices. Since our products heavily depend on these components, the manufacturing cost increased drastically—sometimes even 6–7 times higher than before—making our earlier selling price unsustainable. This affected not just one product but our entire product range.
To handle this, we aligned with the industry-wide situation instead of trying to absorb the losses, as even companies like Raspberry Pi were facing similar price hikes. We adjusted our production and pricing strategy, continued supporting users who want to experiment or build solutions, and planned for scale once the memory market stabilises, which we expect to improve in the near future.
Q. What does your team structure look like?
A. Yes, we have built a completely in-house team with nearly 70 engineers, structured to handle full-stack solution development end-to-end. From the beginning, our vision was to build independent capabilities across every layer of technology rather than relying on outsourcing, because critical areas like driver development require speed, deep integration, and close collaboration. That is why we have dedicated teams for hardware design, firmware development, Linux systems, FPGA, AI compilers, solution engineering, and sales support, all working together under one roof in Surat to deliver comprehensive and integrated solutions efficiently to our customers.
Q. How many units have you sold so far? What revenue did you record in the last financial year?
A. We were still at a very pre-revenue stage in the last financial year because, for the past two years, our focus has been on selling beta kits and onboarding users onto the Vicharax platform. So far, we have sold around 500 Axon units, 100 Vaaman units, and over 5000 Shrike-lite units. With the orders currently in the pipeline and upcoming production plans, our target this year is to sell at least 10,000 Axon/Vaaman units and 50,000 Shrike series units.
Q. How has your funding journey worked? Are you bootstrapped, supported by the government, or backed by investors?
A. Initially, I used my own funds from my earlier consulting company, which had around ₹20–25 million. For about a year after starting the new venture, we were completely bootstrapped and running on that money. When that wasn’t enough, we had to take on some debt and raise small amounts in between to keep the company going, especially around 2023–24 when funding was very limited, and there was no government support.
Later, in late 2025 (around November–December), we raised around ₹45 million from 2–3 angel investors. In addition, we received about ₹4 million from the IIT Kharagpur incubation support. That funding helped us scale manufacturing and develop multiple products—around 6–7 products that each require significant prototyping and initial production costs. We haven’t received direct government funding. The main support has been through incubation and angel investment, and currently, the focus is on scaling manufacturing and expanding the product line.
Q. What are the current challenges you face as a startup in India, and how are you addressing them?
A. The main challenges for startups in India, especially in hardware and order-centric sectors, are limited access to funding and investor bias toward D2C brands. While some startups manage to raise capital, many others struggle to get support because hardware is seen as more complex and less immediately scalable. Another challenge is talent readiness, India has a large pool of engineering graduates, but many need training and hands-on exposure to work effectively in specialised hardware or deep-tech environments.
However, this is also an opportunity. With the right platform and guidance, fresh engineers can quickly become highly capable, as seen in teams that improve significantly within a few months. Additionally, India’s massive base of software developers can be better utilised by transitioning them into hardware and deep-tech development. The core issue is not a lack of talent, but how effectively it is channelled into the right areas to strengthen the startup ecosystem.
Q. How are you selling globally, and how do international customers buy the product online?
A. We sell globally through online channels, and international customers can buy directly from us with a shipping fee of $28, as we ship to most countries worldwide; currently around 30–35% of our users are international, and alongside direct online sales we are also working with distributors—for example, Arcitec in Japan has already sold about 500 units and is planning larger orders in the future—while manufacturing usage is still limited due to mostly manual processes, we expect to expand and integrate more as production scales up.
Q. Do you provide community support for your users?
A. Yes, we provide strong community support for our users. From the beginning, we focused on building an active support system where all our engineering team, around 50 engineers, along with me, are available on our platform to directly answer user questions and resolve issues. We currently have a community of about 1500 members who regularly engage with us, and we respond to their queries as part of our company support. At the same time, we encourage users to interact with each other and not depend only on official responses, as we believe peer-to-peer learning is important for a strong ecosystem.
Q. Are you currently looking for new partners, such as channel partners, vendors, or academic collaborations?
A. Yes, we are actively seeking new partners. For vendors, we are already in ongoing discussions, and we are especially interested in expanding our distribution network by onboarding more channel partners to help deploy our solutions.
We are also open to and actively pursuing more collaborations with academic institutions. We already have partnerships, including with IIT, and we welcome additional tie-ups. Our platforms are designed to support a wide range of educational levels—from school students (like 9th grade using entry-level tools) to advanced researchers and PhD-level users—offering strong technical support across the learning spectrum. Any academic institution interested in enhancing computing and technology education is welcome to collaborate with us.
Q. Are you also providing any videos or learning materials (like manuals or videos) so that a 9th-grade student can use your board and learn more?
A. Yes, we already provide a good amount of learning material, including many videos available on YouTube, and we continuously add more examples, documentation, and video content daily. Along with this, structured courses are also being developed, including a partnership with IIT Gandhinagar, where they are preparing a full course for upcoming engineering students. In addition, several professors and educators are also contributing by offering free video support and educational content related to the platform, making it easier for students, including those in 9th class, to learn and explore more effectively.
Q. What are your future plans, and where are you focusing your investments?
A. We are currently investing heavily in both software and hardware engineering teams, including an FPGA-focused group and experienced engineers. A significant portion of our investment also goes into marketing, sales, and manufacturing. We plan to set up our own manufacturing line in Surat, and the next goal is to scale both production and market reach as our product line expands.
As we grow, we will also scale support teams as needed. On the engineering side, we already have teams across different domains, but we will continue expanding. I also prefer hiring freshers even when not strictly required, because I value the energy and new ideas they bring. It also allows us to support and develop young engineering talent in India by giving them a strong platform to grow.
