AI-Powered Document Ingestion and Data Extraction
APR's advisors coordinate estimates between clients and agencies, and were entering those project documents into their system almost entirely by hand, often after a project had already closed. Ollon built an ingestion pipeline that pulls in PDFs, estimates, and other documents as they arrive, using Gemini to extract line items, vendors, and totals from each file. The extracted data is pushed directly into APR's existing system, and the pipeline has replaced a process that once required a dedicated team, turning data capture into something that happens as projects unfold.
Conversational AI Agent Development
APR's advisors negotiate advertising deals for clients and needed a fast way to pull up prior numbers mid negotiation. Ollon built a chat agent on top of the AlloyDB vector store holding APR's extracted project data, so advisors can ask a direct question and get an answer pulled from real project history. The agent gives frontline staff access to information that used to live only with whoever entered it, folding those insights into their daily workflow.
Custom application development
APR needed a system that could automatically capture and process the estimates and documents flowing in from other advertising agencies as projects happen. Ollon built a new application for APR that integrates with its existing systems while adding an automated ingestion and analysis layer, working alongside APR's own development team on the API updates needed to push data back into the platform advisors already use. Advisors now see estimates and documents processed into the platform as they arrive.
Data and Analytics
APR, an advertising agency, had most of its project and deal information arriving as unstructured PDFs, so Looker's reporting only reflected the smaller set of data that already lived in a database. Ollon's document extraction pipeline changes that, pulling structured data out of those documents and pushing it directly into APR's data warehouse. This enabled APR's team to build reports in Looker that draw on deal information no dashboard could reach before.
Unstructured data processing
APR, an advertising agency, receives estimates and project documents in a wide range of formats from different agencies, none of it structured for easy entry into a database. Ollon's team initially planned to use conventional OCR to detect document sections, then pivoted once large language models proved capable of parsing the documents directly. That shift meant the system Ollon built could extract meaningful data directly from PDFs and other unstructured document types APR receives.
Google Cloud
APR needed a system that could capture estimates and bids as they arrive. Ollon runs APR's document ingestion platform entirely on Google Cloud, moving uploaded files through a serverless pipeline built on Cloud Run and Cloud Functions where Gemini extracts structured data and generates embeddings for search, all stored in AlloyDB, Google's managed PostgreSQL database, which also serves as the vector store the AI chat agent queries. The pipeline scales automatically as APR's document volume grows, with no servers to provision.
Phased rollout planning
APR wanted to replace legacy systems and workflows with AI, but AI could only add value once the underlying data existed and had been validated. Ollon phased the rollout accordingly, starting with a first phase that centralized and validated APR's project data through the ingestion pipeline, then a second phase that layered the chat agent and analytics on top of that validated data. Further phases, covering new features and deeper workflow replacement, are being planned now. This let APR gain business value with every release, while giving both teams a feedback loop to adjust priorities as the business built confidence in what AI could do.
