Hiding in plain sight
We always come back to the basics. The principles of information management in the cultural domain and the world of technologies are surprisingly similar: the world serves as a stage for the contemporary engineers of the human soul, who seek to turn men and women into predictable protein-based intelligent products.
While educational, religious, and artistic works have long been used to shape human behavior, cinema has emerged as a particularly powerful medium for this purpose. Consequently, many Hollywood movies prioritize partisan political programming: portraying the inherent goodness of collectivism in Star Trek; highlighting the ultimate evil of corporations in Avatar and The Meg; and depicting military code of honor as institutionalised corruption in A Few Good Men.
Unless one is a discerning moviegoer and a history buff, sensing the cultural bias in films is a complex task. The same issue exists when trying to figure out changes across hundreds of pages of a new version of engineering documentation, especially in PDF format.
Although these documents are transmittal-friendly, comprehensive repositories of essential data, their siloed and static nature presents a significant concern for their shop floor consumers. With each new version, seemingly minor updates can have far-reaching impacts that compromise efficiency and safety. Therefore, continuously capturing and structuring data such as part numbers, tolerances, sequence of steps, and even the chemical composition of materials into BOM-like structures is paramount to a company’s bottom line.
The Senticore solution uses an AI-led, human-assisted method of PDF reconstruction. Output varies: an MS Word work plan for humans to read, or JSON for application APIs to ingest. We chose the human-readable use case as our strategic go-to-market choice simply because it is easier to demonstrate to our target aerospace and defense audience.
The combination of generative AI, specialized neural networks, and graphs can detect semantic and visual changes between documentation versions and alert end-users. Furthermore, using semantic identifiers and other key artifacts as logical connectors makes it easy to integrate the solution with other engineering systems via API. This is the digital equivalent of watching a movie and instantly understanding all the details, overt or covert cultural references, and mind traps.
One client’s feedback highlighted the product’s ability to reduce manual work by up to 90%. Another estimated that it can save them at least 30 hours for each work plan, providing a significant boost to their profit margins. Crucially, the same techniques can be applied to migrations from legacy text-based MES systems to their more modern, model-oriented incarnations.
A technical challenge we are currently focused on is optimizing the speed of the initial AI extraction and annotation process for PDF documents while managing compute costs. The solution distributes the algorithmic load across a pool of inexpensive, on-demand GPU compute. Interestingly, we have observed that the AWS GPU offering, while five times more expensive than those from comparable small cloud vendors, translates directly into a better SLA, often required for mission-critical deployment.
If you are running an engineering and manufacturing enterprise and have a technical documentation gap in the digital thread, we can help you bridge it, allowing your business to run more efficiently. This includes catching unusual changes or previously unexplored plot twists in the documentation versions early, preventing unpleasant surprises. The opportunity to discuss cultural trends and movies over a good bourbon comes as a bonus.
