Insights

We are currently working on a new assistant focused on value investing support using U.S. SEC EDGAR 10-K filings.

The solution combines structured 10-K parsing with complex Retrieval-Augmented Generation (RAG) techniques so analysts can explore company fundamentals with more context and traceable evidence.

The pipeline automatically downloads 10-K reports from EDGAR and uses a caching layer to speed up repeated analysis requests and improve processing efficiency.

The assistant helps users save time and effort in analysis, while the final investment decision is always made by a human.

At this stage, we are validating document ingestion quality, retrieval relevance, and answer grounding before broader release.

We have launched another closed beta.

This time, we focus on a problem many investors face but rarely see addressed properly: the real profitability of algorithmic trading managed by external companies.

In many cases, external traders or signal providers are incentivized to artificially boost reported performance by highlighting winning trades while ignoring transfers, capital changes, or hidden losses. As a result, reported performance can look excellent, while actual investment performance remains far less impressive.

Our solution connects to a trading account through a Read-Only (RO) API key, analyzes complete account activity, and provides a clear, unbiased report of true performance.

Instead of relying on selective performance highlights, we evaluate the full account history so investors can see the real picture across traditional and digital assets.

If you are working with external traders, funds, or signal providers and want to understand how your portfolio is really performing, this solution is for you.

Interested in testing the closed beta? Please contact us directly, and we will be happy to share more details.

We have created two test clients for our business assistant: one for iOS and one for Android.

Both clients are currently in active testing and are not yet published on the Apple App Store or Google Play.

At this stage, we can share access with a limited number of selected customers for early validation and feedback.

We are pleased to announce the launch of a closed beta of our visualization platform designed for construction companies building residential properties and housing projects.

Our solution enables teams to generate high-quality exterior and interior visualizations before construction begins, based on real materials, architectural layouts, and planning documentation.

We are currently inviting a limited number of construction companies, property developers, and real estate teams to participate in the beta program.

If you are interested in early access, please contact us directly to learn more.

As part of our ongoing work on scalable AI and GenAI infrastructure, we deployed and tested NVIDIA Triton Inference Server on Google Cloud using Debian-based T4 GPU instances.

The objective was to evaluate Triton as a potential serving layer for production-grade model deployment, with a focus on performance, GPU utilization, operational complexity, and cost efficiency.

Our testing covered GPU-backed model serving on GCP, containerized Triton deployment, model repository configuration and versioning, inference latency and throughput behavior, and resource utilization with cost considerations.

Overall, the evaluation confirmed that Triton provides a flexible, production-ready serving architecture for GPU workloads. It also offers strong support for multi-model and scalable deployment patterns. While setup complexity is manageable, production usage requires careful configuration and operational discipline.

For experimentation and moderate-scale inference, T4 GPUs proved to be a cost-effective baseline.

This work informs our infrastructure decisions for AI-enabled products, including document processing and LLM-based applications.

We delivered a number of new features for our business assistant project.

In parallel, we updated our product map to include many additional use cases based on ongoing customer needs and validation insights.

This update gives us a stronger foundation for upcoming iterations and broader adoption across business workflows.

We are happy to share that we graduated from the INSEAD AI Venture Lab Acceleration Program.

Throughout the program, we gained valuable insights from guest speakers and the invited expert team, covering product strategy, venture execution, and scaling decisions.

The experience helped us sharpen our direction and improve how we build and deliver our AI products.

We launched our product with the first core business assistant functionality: correlating bank transactions with invoices or contracts.

This release is focused on helping teams reduce manual reconciliation effort and improve financial process clarity from day one.

We are proud to share that we were accepted into the FIRST cohort of the AI Venture Lab Acceleration Program.

The program gives us access to expert mentorship, structured venture support, and a strong network of founders and industry leaders.

This milestone strengthens our product and go-to-market execution as we continue building and validating our AI solutions.

We released a new book to support candidates preparing for the Google Cloud Professional Cloud Architect exam.

The book includes two sets of complex practice questions and covers all major areas required for the exam.

Based on the early interest in the book, we are also assessing an opportunity to build a dedicated subscription-based exam-prep web application to help customers prepare more effectively.

At the beginning of June, we launched our community project to help people prepare for the naturalization test.

Many available mobile apps, including those on Android and iOS, are heavily commercialized and can be distracting during learning.

We built our app to be suitable across different platforms and to provide a focused preparation experience without ads.

At the end of May, we completed the Startup School online course from Y Combinator.

The program gave us practical guidance on startup execution, customer focus, and company building discipline.

At the beginning of May, we prepared a full package of application materials and submitted our application to Y Combinator.

The preparation process helped us sharpen our positioning, clarify key milestones, and strengthen our product and execution narrative.

At the beginning of April, we introduced a Telegram bot to support end users with algorithmic trading questions and operational tasks.

The bot is now integrated with our backend and helps streamline day-to-day solution operations with faster user guidance and more consistent execution support.

This addition improves accessibility for users who need quick assistance while maintaining a reliable connection to the core trading workflow.

We released a Telegram bot that allows users to run trading account audits directly from a chat interface.

The bot connects to broker and exchange accounts through Read-Only API keys, retrieves full trading history, and generates key performance metrics including Sharpe ratio, Sortino ratio, and risk-reward analysis.

This makes it easier for users to request and receive account performance reports without needing access to a web dashboard or desktop environment.

We completed end-to-end testing of the trading account audit solution across multiple exchange accounts and trading scenarios.

Testing covered a wide range of account types, trading patterns, and edge cases to ensure accuracy and reliability of performance metrics before the planned closed beta launch.

In parallel, we began preparing the Telegram bot interface to make audit functionality accessible directly from a chat environment.

We completed the reporting layer for the trading account audit solution, producing structured performance reports that clearly present key findings to end users.

Reports now include visual summaries of portfolio performance over time, risk-adjusted return metrics, drawdown analysis, and benchmark comparisons, presented in a format accessible to non-technical investors.

We added benchmark comparison capabilities to the trading account audit solution, allowing users to compare their managed account performance against simple buy-and-hold strategies for the same assets and time period.

This feature highlights whether active management is adding value compared to passive holding, which is one of the most important questions for investors evaluating external traders.

We completed the implementation of a comprehensive risk metrics engine for the trading account audit solution.

The engine calculates key portfolio performance indicators including Sharpe ratio, Sortino ratio, risk-reward ratio, maximum drawdown, and win rate across configurable time windows.

We expanded exchange coverage and improved the reliability of our trading account audit data pipeline.

The solution now handles edge cases in trade history more accurately, including partial fills, internal transfers between sub-accounts, and fee calculations across different exchange fee structures.

We also improved how capital inflows and outflows are tracked, ensuring that performance metrics reflect actual investment results rather than gross trading volume.

We built a data pipeline to normalize trade history from digital asset exchange APIs into a unified internal format.

Different exchanges report trades, deposits, withdrawals, and transfers in significantly different structures and conventions. Our pipeline handles these differences and produces a consistent dataset suitable for accurate performance analysis.

This step was essential to ensure that audit results are comparable across exchanges and account types.

We completed the initial integration with major digital asset exchange APIs for account data collection using Read-Only API keys.

The integration supports retrieving full trade history, balance snapshots, deposit and withdrawal records, and fee information. All connections use Read-Only access, ensuring no trading or fund movement is possible through the audit process.

We completed the requirements analysis and system architecture design for the trading account audit solution.

The goal was defined: provide investors with an unbiased, transparent view of their managed trading account performance by analyzing full account activity rather than relying on selective reporting from external traders.

We designed the system architecture to support secure API connectivity, scalable data processing, and modular metric calculation.

We developed an initial proof of concept for the trading account audit tool, validating the core idea of connecting to exchange accounts via Read-Only API keys and computing basic performance metrics.

The prototype confirmed that sufficient data is available through standard exchange APIs to calculate meaningful risk-adjusted return indicators and detect discrepancies between reported and actual performance.

We conducted research into established portfolio performance measurement methodologies and their applicability to algorithmic trading accounts.

The research covered risk-adjusted return metrics commonly used in traditional finance, including Sharpe ratio, Sortino ratio, Calmar ratio, and maximum drawdown analysis, and evaluated how these apply across traditional and digital asset markets with different liquidity and trading-hour profiles.

This research informed the design decisions for our trading account audit solution.

We identified and analyzed a recurring problem in managed algorithmic trading: the lack of transparent, independent performance reporting.

Many external traders, signal providers, and fund managers report performance selectively, highlighting winning trades while omitting capital changes, fees, or losing periods. This creates a significant gap between reported and actual investment performance.

We began exploring how an independent audit tool could address this transparency gap for investors across traditional and digital assets.

We began exploring opportunities in trading account analytics, focusing on the growing need for reliable performance tracking and verification tools.

Initial work included evaluating available broker and exchange APIs, understanding data access patterns, and assessing the feasibility of building automated account analysis capabilities.

This exploration laid the groundwork for the trading account audit solution developed in the following years.