Unboxing LLMs > loading...
Systems Thinking, Real-World Results.
R&D, AI, Infra — One Project at a Time.
Bridges, Bots, and Billion-Dollar Pipelines.
Built MediForecast to monitor realtime status of diseases, medicines, and shortages, delivering competitive intelligence for pharmaceutical companies.
Powered by an LLM-based multi-agent data transformation pipeline and a NextJS frontend, it provided actionable analytics to industry stakeholders.
Designed Transformator.AI to provide data transformation endpoints for on-the-fly ETL operations on structured and unstructured data.
Achieving top performance in parsing English medical data and SWIFT transactions, it streamlined workflows and improved data management efficiency.
Developed STOXFlow, a machine learning-driven platform for real-time detection of chart and harmonic patterns across financial markets.
Offering actionable statistics, it enhanced trading decisions and was later integrated into proprietary trading platforms, showcasing its value in financial technology.
Initiated Super GPTs as a hobby project to maintain my prompt engineering expertise.
By creating custom GPTs on the OpenAI GPT Store, I explored advanced applications of large language models, reinforcing my skills in AI-driven natural language processing and generation.
PS: I even ran a newsletter in AI space by the name of Synapse Circuit back in 2024!
As CTO of AxisCards, I oversaw the development of a next-generation merchant services platform, encompassing POS, web, and API solutions.
I secured partnerships with Google, Apple, PAX, and WorldPay, and drove a company-wide technology shift, enhancing payment services, particularly for pharmacies via IOCSave in Ireland.
https://www.youtube.com/watch?v=fYByFJ9TzbE
Built a local stateful .NET desktop application for Detroit Street Partners, a hedge fund, to monitor and execute trades based on proprietary strategies.
With real-time integration to RealTick and Bloomberg, the app offered dynamic charts and tabular views, later becoming an external offering due to its success.
As VP at ESW, I spearheaded the design and implementation of a cost optimization framework aimed at controlling the enterprise’s growing cloud expenses.
Guided by senior leadership, I introduced automated policy enforcements, real-time usage monitoring, and algorithmic scaling. The success of the framework not only saved millions but was also spun off into two separate product offerings: CloudFix and Virtasant, both focusing on proactive cloud cost intelligence and resource management for global clientele.
I worked with brands like Aurea, Ignite Technologies, Think3, and Trilogy and ran automated cloud optimization on AWS for over 250 managed products.
Swiss Post needed to update its legacy .NET insurance document processing system to improve speed and accuracy.
I led the modernization effort by integrating Tesseract OCR, enabling faster text extraction from scanned forms. This included implementing a new classification module to route documents automatically to the correct processing queue.
The project significantly reduced manual review time, decreased errors, and improved the overall throughput of insurance documentation workflows.
Working with Loop 365 in Ireland, I spearheaded the development of PerSav, a platform designed to deliver optimized healthcare for elderly.
The system ingests user data and applies predictive analytics to suggest optimal tasks, routes, locations and healthcare while enabling timely notifications and monitoring by family members.
Utilizing secure data-sharing protocols, PerSav ensures compliance with local and EU regulations, and provides actionable insights in real time. The solution helped users achieve better control over elderly care.
ReScreen.IO is a social media moderation platform that integrates seamlessly with Twitter. Users can create, share, and subscribe to custom block lists, effectively filtering out abusive or unwanted accounts.
In addition to user-curated lists, I implemented machine learning models—combining k-Nearest Neighbors and Convolutional Neural Networks—to detect threatening, profane, or spam content in tweets.
This approach drastically reduced exposure to harmful language and fostered safer online discourse. The platform’s intuitive interface allows both private and public curation, empowering communities to collectively combat cyberbullying.
Finomial required a robust solution to map and validate customer and transaction data against global compliance standards while integrating with different backend systems.
With Philip Romanik, we architected a dynamic data mapping platform that allowed real-time configuration of data fields, transformations, and validations. By modularizing the logic, the platform could adapt to new regulations and evolving data formats rapidly.
The end result significantly reduced manual intervention and increased data accuracy, which was critical for compliance reporting and audit readiness of hedge fund admins like US Bank, BNP Paribas, and others.
This project focused on creating a symbolic regression engine capable of ingesting numerical data series and outputting candidate mathematical equations that explain the observed patterns.
By applying heuristic and evolutionary algorithms, the system automatically evolved potential models, pruning them based on error metrics and computational complexity.
This helped researchers and data scientists derive human-readable formulas without manually testing endless possibilities, greatly accelerating analytical workflows in fields such as finance, physics, and bioinformatics.
In early 2012, I collaborated with a research team working under a DARPA grant to study and prototype portable, rapidly deployable bridges.
My primary contribution was developing detailed physics simulations in AutoCAD and MATLAB. These simulations modeled critical factors such as tension, torque, stress, and load-bearing capacity.
We validated the simulations by correlating them with real-world measurements, enabling quick iteration and refinement of suitable materials and support structures.
The final outcome helped the team identify the best materials and beam designs for various configurations, ultimately improving the reliability and efficiency of bridge deployment under challenging conditions.
