Parts AI

Inspiration

My journey began a few years ago, fueled by one question: how can AI solve real, high-stakes problems? When I looked across industries, nothing felt more urgent—or more broken—than the supply chain. COVID-19 didn’t just expose its fragility; it cracked it wide open. What I thought was a temporary disruption revealed a deeper, ongoing crisis driven by tariffs, geopolitics, and global dependencies.

This wasn’t just about toilet paper shortages—it was about engineers unable to build, factories sitting idle, and companies hemorrhaging time and money. Today’s sourcing process is slow, manual, and often unreliable. Engineers struggle to find replacement parts, especially when dealing with EOL components or shortages. This leads to production delays, missed deliveries, and customer dissatisfaction. The main issues? Unavailable parts, difficulty finding alternates, lack of supplier visibility, and a manual process that just doesn’t scale. I saw the chance to do more than just analyze data—I could give engineers real-time intelligence to navigate chaos, reduce downtime, and find new paths forward.

That’s why I built Parts AI. This isn’t a side project. It’s a mission to turn crisis into opportunity. I'm using AI to make the sourcing process faster, smarter, and more resilient—because solving this one piece of the supply chain puzzle could unlock innovation everywhere else.

What it does

Parts AI is an AI-powered tool that automates and accelerates the electronic parts replacement process for engineers and sourcing teams. It delivers four key capabilities:

• Real-time visibility into part availability, pricing, and lead times across approved distributors
• AI-powered replacement recommendations based on design criteria
• Automated supplier matching with sourcing preferences built-in
• Natural Language Processing (NLP) that eliminates the need for manual, time-consuming searches

With Parts AI, engineers can upload a Bill Of Materials (BOM) and instantly get intelligent, ranked suggestions—turning what once took hours into a matter of seconds.

How I built it

I began by defining the key requirements and MVP scope, focusing on automating the parts replacement process with real-time sourcing intelligence. I used Lovable to build the frontend, which handled BOM file uploads, processing, and displaying structured BOM contents to users.

For the backend, I created AI agents using Toolhouse AI, equipping them with tools for parsing, matching, and ranking parts. I also integrated external distributor APIs and provided access to those APIs through Toolhouse, enabling the agents to fetch real-time availability, pricing, and lead time data.

Additionally, I built a natural language agent using Toolhouse that allows users to ask freeform questions about sourcing, availability, or replacements. This agent intelligently calls the underlying tools and APIs to generate accurate, contextual answers—making the platform even more intuitive and powerful.

These agents were deployed as APIs that seamlessly power the backend logic, allowing the frontend to deliver fast and intelligent part recommendations.

Challenges I ran into

Integrating real-time data into the Lovable frontend was a significant challenge. Uploading and parsing the BOM required calling distributor APIs on-the-fly to fetch current availability and pricing data. Lovable could not handle it and I was not getting any data back. I resolved this by offloading API calls to a Toolhouse agent, which handled the logic and passed back clean, structured data for Lovable to render effectively.

Adding natural language capabilities was another complex task. I implemented agent chaining in Toolhouse to allow one agent to interpret the user's intent and another to execute the corresponding API calls. This architecture allowed for flexible, intelligent responses but took several iterations to get right.

I also encountered stability issues with the Lovable platform—making UI changes would occasionally break existing functionality, which required careful testing and frequent rollbacks. It was especially hard to get Lovable to consistently display results returned by the Toolhouse agent. It was also hard to debug Lovable app to see why it wasn't displaying the full set of results. Additionally, I ran into API rate limits from distributors, which temporarily throttled my ability to test real-time data integrations and slowed down iteration speed.

I also explored integrating Arize for model evaluation and Groq for ultra-fast cloud inference. However, as of Saturday night, those services were not compatible with each other in this stack, so I had to defer those enhancements.

Despite these challenges, each hurdle pushed the solution toward greater robustness and usability.

Accomplishments That I Am Proud Of

I'm proud of taking an idea from scratch to a fully working product in just a weekend—powered entirely by the latest AI tools. I was able to go from concept to execution using a combination of Lovable, Toolhouse, and real-time distributor APIs to solve a real-world sourcing problem for engineers.

The product includes an interactive frontend, AI agents with real-time data capabilities, and even a natural language agent that answers user queries—all connected and working in a live demo. This project demonstrates how AI can be applied not just for ideation, but for building and shipping high-impact, technically sound products at speed. The fact that the system already shows measurable time savings and real user value makes it even more rewarding.

What I learned

This project taught me just how powerful and accessible modern AI tools have become. I learned how to go from problem definition to a functional, end-to-end product in record time by leveraging platforms like Toolhouse, Lovable, and external APIs. I deepened my understanding of how to build and use AI agents effectively, work around platform limitations, and integrate real-time data sources—all while keeping the user experience smooth and reliable.

I also learned the value of iteration under pressure. Whether it was debugging UI issues in Lovable, hitting API rate limits, learning how to build AI agents using Toolhouse, every obstacle pushed me to become more resourceful. Although Arize and Groq did not support integrations with Toolhouse, I learned about their capabilities and the value the offer. I look forward to using them in future. Most importantly, I saw firsthand how AI can turn a traditionally slow, manual process into something fast, intelligent, and scalable.

What's next for Parts AI

Parts AI has the potential to become a core workflow tool for engineers and sourcing teams in electronics manufacturing. The next steps include building a team, expanding support for more distributor APIs, improving part-matching accuracy with richer component datasets, and refining the UI for enterprise-grade usage. I'm also planning to integrate observability tools like Arize and faster inference engines like Groq to make the platform even more robust and responsive.

In parallel, I’ll be conducting user interviews with sourcing professionals and design engineers to validate the product-market fit and prioritize features for an early pilot. The goal is to evolve Parts AI into a trusted, go-to platform that helps companies reduce downtime, improve sourcing resilience, and streamline hardware development in an increasingly volatile supply chain environment.

Built With

• distributorapis
• lovable
• toolhouse