We developed a full-stack data tool and took our client’s vision to a production-ready AI solution
Custom AI system for tracking grants
MassVentures is a Boston-based VC firm that had a need for a tool which would help them track government grant opportunities. They approached us with a concept and it was our job to turn it into a fully operational AI tool that’s intuitive, useful, and ready to go.
In data science, you need to become a domain expert really fast. That’s exactly what we did to bring our client’s idea to life. And when you're working with AI models, clarity matters. We focused not just on the tech, but on making sure the output was easy to understand and actually useful.
What we did
Data engineering
AI
Backend
Solution Architecture
Frontend
UX/UI design
Projected grant revenue increase per year
We built a full-stack data tool that unifies scattered workflows into one platform, giving clients everything from research to planning in a single dashboard. It pulls data from multiple sources, scrapes news and social content, and visualizes trends clearly for easy tracking and analysis.
Since off-the-shelf models didn’t cut it, we built a custom AI system from the ground up. We embedded texts into vector space using a neural network, used our own heuristics to group similar items, and ran an LLM to define topics. It’s tailored specifically for tracking government grant content. And it works.
Strategic value and insight per year
We used Airflow, AWS Fargate, Scrapy, and LLMs to automate and scale core processes. Scrapy handled extraction, while Airflow triggered on-demand Fargate tasks for Python jobs letting us process large volumes efficiently without overbuilding infrastructure. On the backend, FastAPI, SQLAlchemy, Pydantic, and Alembic kept development fast, clean, and fully Python-based from models to deployment.
We shipped a production-ready AI tool that brings together data pipelines, backend apps, and cloud deployment. It handles everything from variable text lengths to topic prediction, and it’s built to deliver real insights, not just outputs. Fully integrated and fully operational.
Working on a full stack data projects is always a thrill. From mapping client requirements to discussing with UI/UX team how the necessary data is presented, its always dynamic project with many moving pieces. What made this project even more fun, was the necessity to develop our noval AI approach for topic modeling. We tested what was available out there, but nothing made the mark. We had to get our hands dirty.
Domjan Barić
Head of Data Science
Discover more projects
We developed a full-stack data tool and took our client’s vision to a production-ready AI solution
Custom AI system for tracking grants
MassVentures is a Boston-based VC firm that had a need for a tool which would help them track government grant opportunities. They approached us with a concept and it was our job to turn it into a fully operational AI tool that’s intuitive, useful, and ready to go.
In data science, you need to become a domain expert really fast. That’s exactly what we did to bring our client’s idea to life. And when you're working with AI models, clarity matters. We focused not just on the tech, but on making sure the output was easy to understand and actually useful.
What we did
Data engineering
AI
Backend
Solution Architecture
Frontend
UX/UI design
Projected grant revenue increase per year
We built a full-stack data tool that unifies scattered workflows into one platform, giving clients everything from research to planning in a single dashboard. It pulls data from multiple sources, scrapes news and social content, and visualizes trends clearly for easy tracking and analysis.
Since off-the-shelf models didn’t cut it, we built a custom AI system from the ground up. We embedded texts into vector space using a neural network, used our own heuristics to group similar items, and ran an LLM to define topics. It’s tailored specifically for tracking government grant content. And it works.
Strategic value and insight per year
We used Airflow, AWS Fargate, Scrapy, and LLMs to automate and scale core processes. Scrapy handled extraction, while Airflow triggered on-demand Fargate tasks for Python jobs letting us process large volumes efficiently without overbuilding infrastructure. On the backend, FastAPI, SQLAlchemy, Pydantic, and Alembic kept development fast, clean, and fully Python-based from models to deployment.
We shipped a production-ready AI tool that brings together data pipelines, backend apps, and cloud deployment. It handles everything from variable text lengths to topic prediction, and it’s built to deliver real insights, not just outputs. Fully integrated and fully operational.
Working on a full stack data projects is always a thrill. From mapping client requirements to discussing with UI/UX team how the necessary data is presented, its always dynamic project with many moving pieces. What made this project even more fun, was the necessity to develop our noval AI approach for topic modeling. We tested what was available out there, but nothing made the mark. We had to get our hands dirty.
Domjan Barić
Head of Data Science
