Meet EJO -Aftermarket Parts Assistant Bot
The Answer To "What year is E150FCXSSM?" and "Is this prop compatible with my transom?"
What part fits the engine model serial number sequence when the only reference is print manuals and all integrations are gated?
My father dedicated decades to OMC before it was acquired by BRP, and in 2020, BRP announced the discontinuation of Evinrude-Johnson outboard engine manufacturing, reducing the number of American manufacturers competing with an increasing number of Chinese and Japanese-owned companies. When production of parts was confirmed to continue, and based on feedback from dealerships I had worked with, I recognized the impending communication bottleneck that would affect all dealerships servicing Evinrude-Johnson engines once manufacturer support ceased entirely. This realization inspired me to create a chatbot assistant for dealerships and mechanics providing repair and service for Evinrude-Johnson models sold prior to their sundowning in 2005, as these models can typically only be referenced from manufacturer catalogs from that year, many of which are only in print. As my father and I contemplated selling parts from his garage, I understood that matching parts would be the most challenging aspect of the project, leading me to pursue EJO, which began with training Anthropic's Fable model on relevant literature to build the knowledge base now available in this demo. EJO was intentionally designed not to rely on LLM APIs due to their tendency to provide inconsistent and incorrect information; instead, it harnesses robust coding for efficiency and resource optimization.
Currently, EJO is capable of deployment across virtually any server and database, seamlessly integrating into most websites without requiring changes to their existing frameworks. Future updates for EJO will include support for additional manufacturer catalogs and expanded capabilities for various models, engine specifications, and potentially other industries.
What makes EJO better then just typing the serial number into an AI?
When I was a kid, I grew up with chatbots like AOL's "SmarterChild". These simple, legacy Perl, precursors to the modern AI chatbot were pretty good all by themselves, without the transformer neural network or ability to scrape the web for pseudointellectual follow up responses. If you want to experience the nostalgia, you can click the link to visit a live simulation of what it was like at that time (if you didn't happen to be around or know). SmarterChild was pre-loaded with parsed 'answers' to user messenger responses, like "What does the weather look like in San Francisco?" or "What is the biggest animal in the world?"
This was encoded by the person writing the chatbot and carefully engineered to parse and recognize (with appropriate follow up) thousands of phrases. Its 'memory' was non-existent beyond pre-cached scripts that would respond to the user when the user opened up the conversation in AOL Instant Messenger.
EJO was made to perform a single purpsoe and perform that purpose well- not to expand beyond that. Its training was rooted in retrieving archive and print data and parsing language triggers that would be asked by a regular customer. Asking the same of a current AI assisted search engine or chatbot will not give the same accurate detail where it matters most. In the future, as people run their engines and require parts to repair and service them, EJO can potentially help logistically, as OMC era parts, related non-OEM parts, and BRP Evinrude-Johnson parts start to fade on the market.
EJO is not much different than SmarterChild, and this was intentional - not from lack of technical knowhow or other constraints.
Where LLMs and transformers makes sense in several usage cases (or so we think), in this particular case, a scripted chatbot would not answer the user with hallucinated facts, then transcribe them in an authoratative tone.
What are the technical specs of EJO?
EJO was designed using vanilla javascript to be lightweight, easy to install, and not impact the frontend of dealership web pages that may be running legacy systems. Its code has been debugged and tested against common plugins to ensure minimal conflict. At its core, its grammar-language parser does the heavy lifting when it comes to retrieval, and the necessary constraints that are already in place using vanilla Javascript mean that it will not mislead the customer and ship the wrong part for the model. EJO can assist with identifying propellors, machine parts, model, and year for most Evinrude-Johnson models and plans to expand its understanding of other manufacturer's legacy models are its main potential for growth. In the future, EJO may be enhanced using AI to assist with repairs and service of legacy outboard engines, particularly by people living in a rural area whose lives depend on keeping that engine running, especially once the availability of compatible parts starts to become scarce.
Browse Other Projects & Blog Posts
