top of page
Search
  • Writer's picturePredii

Introducing: AI-powered part search with Predii 360™ Platform

Updated: May 28

At Predii, our purpose is clear: we firmly believe that AI can be a powerful ally for every player in the automotive parts and service ecosystem. We envision a future where AI augments human expertise and seamlessly integrates with workflows that enhance efficiency, productivity, and increase revenue across the board. 


Today, we're happy to announce the beta release of our new part search in Predii 360™, a Predii LLM-powered AI platform designed to make parts look-up faster, and more accurate.


Predii 360 assists workflows that involve anything from a service advisor assisting a car owner with their vehicle service, communicating with a technician about a customer's repair estimate, to a technician requesting parts from either a parts counter or a parts warehouse distributor, to seeking expert opinions on repairs.


“With every vehicle serviced, it's like the clock's ticking and quoting each part for every job eats up precious time. The struggle is finding that sweet spot between getting it right and getting it done. I imagine AI would be like having an extra smart, tech-savvy assistant in the garage." - Stevie Whitlow, Engine Technician, Madison, TN

How Predii 360 powers part look-up


The Predii 360 part search capability is built on the hybrid model of our Predii 360 platform. Predii LLM, Predii’s specialized large language model for part and service supports automotive specific intent discovery.


Domain-focused workflows - like part lookup - require specialized knowledge sources that power RAG/OAG pipelines in Automotive domain.


"I need a BOO switch for a 2020 Ford F-150"

Imagine this scenario: A technician is trying to order a brake light switch for a 2020 Ford F-150. Instead of using the catalog part name, the tech types in a search request for 'BOO switch' - jargon for Brake On/Off switch. Predii 360 processes the query and identifies the intent. It uses vector embeddings of automotive ontology to convert terminology from a technician service world to parts catalog terminology. Then, Predii 360 performs a semantic retrieval and then re-ranks the results and identifies the correct term ('brake light switch'). The system queries the Parts Catalog API with the corrected term, obtains relevant results, and presents the matching part back to the tech.


Specialization involves all steps of the pipeline: Intent discovery to Intent task execution to retrieval to reranking of retrieved results.


Predii 360 sets industry standard in high-precision retrieval of parts information.


In parts and service, ‘Right Part, Right Place, Right Time' is the cornerstone of operational efficiency and customer satisfaction. Both speed and accuracy in retrieving correct part information are crucial - and need to be reflected in the 'precision’ with which supporting AI solutions interpret and respond to user queries.


Our part search capability in Predii 360 has been rigorously evaluated on a dataset of more than 10,000 generated instances, with a wide variety of input variables including things like user typos and non-standardized part terms. Predii 360 beta currently achieves an impressive precision rate of 96%, showcasing the ability to identify the correct parts while handling real-world variables and uncertainties in the parts lookup process.


Predii 360 achieves 96% Accuracy in retrieving the correct part

The value of high precision in the parts lookup and ordering process cannot be overstated. For businesses, it translates to significant reductions in time and resources spent on correcting orders, leading to cost savings and increased operational efficiency. For customers, it means faster service times and the assurance that their vehicles are being serviced with the exact required parts, enhancing trust and satisfaction.


Understanding the 96% Precision Rate

A precision rate of 96% indicates that in 96 out of every 100 queries processed by Predii 360, the system correctly identifies and suggests the relevant part terminology based on the user’s input. Let’s try to understand the precision rate stage-by-stage.


  1. Query Processing and Intent Identification: Predii LLM accurately processes user queries, identifying the intent and extracting crucial details with a high degree of accuracy. A 96% precision rate here means the model correctly understands and categorizes the vast majority of queries, setting a strong foundation for accurate parts lookup.

  2. Vector Embedding and Similarity Search: By translating part names into vector embeddings and performing a similarity search against a vector database, the system effectively narrows down the vast pool of potential parts to the most relevant options. The precision rate suggests that these embeddings and searches are highly effective, even in the face of ambiguous or complex queries.

  3. Re-ranking and Selection: The re-ranking process, which narrows the results further to the top 5 selections, and the subsequent selection of the highest-scoring result based on the longest common subsequence, are critical for enhancing precision. The achievement of a 96% precision rate demonstrates the ability to consistently deliver precise and relevant part suggestions, minimizing errors and enhancing the efficiency of parts lookup.

  4. Integration with part catalog API: The final query to the part catalog API (currently integrated with eBay for demo purposes), based on the normalized part name, and the processing of results by Predii LLM to generate the final output, are the culmination of the pipeline. The high precision rate here demonstrates the seamless effectiveness of the entire process, from understanding the initial query to providing actionable, accurate suggestions that meet the user's needs.


How Predii 360 solves challenges in part search

This is the vision: as soon as a vehicle is on file with a VIN and a diagnosis, an AI-powered platform will be able to respond with everything there is to know about this particular repair: parts, repair procedures, tutorials, VIN and mileage-specific related repairs and suggested maintenance, safety recalls, etc.


Predii 360 supports this with three core capabilities:


Intent Discovery powered by Predii LLM

Predii Intelligent Intent Response Network (IIRN) reliably discovers the intent behind part related questions including typos, OEM specific terminology, and technician ‘service’ language.

Example prompt: "I need a BOO switch for my 2020 Ford F-150."











Part & Service specific RAG/OAG and vector database

Predii 360 domain-specific RAG/OAG architecture connects relevant knowledge sources embedded in the vector database, effectively retrieving interconnected intelligence based on user prompts in any point in the workflow.


Example prompt: "Show me what I need for a brake job on a 2019 Ford F-150."


Predii 360 effectively interprets the intent of brake job and tying it back to related parts for the brake job, related labor times for performing a brake job, related safety concerns (TSBs, recalls) around brake systems, etc.





Intent chaining powered by Predii LLM (beta)

Automotive-specific intent chaining actively guides users through a sequence of VIN-specific related events, i.e. from symptom to diagnosis, to part identification, to repair instructions.


Example prompt: "Great! Now show me how to remove and install brake pads on a 2019 Ford F-150."










Test Predii 360™ Parts Co-Pilot beta





 

About Predii 360™

Predii 360 is our full-scale Generative AI Solution for Automotive Service data, built to understand and intelligently respond to automotive repair and service questions. A hybrid approach combines state-of-the-art AI with over a decade of Predii's domain expertise in parts & service. Predii 360 unlocks all the power and convenience of Generative AI while solving major concerns around reliability (“hallucinations”), IP, and data security.


Learn more about Predii 360 on our website or sign up to test the beta version at predii360.ai

218 views0 comments

Comments


bottom of page