Behind the Tech: How We're Automating Systematic Literature Reviews
We’re going Behind the Tech to spotlight the team behind our integrated approach, combining strategic consulting with purpose-built technology to help clients plan and execute projects with speed, scale and precision.
In this Q&A, we’re featuring the developers behind our systematic review platform (ANCHOR). From automating literature screening to structuring complex data from PDFs, the team is focused on helping users move from messy documents to clean, comparable insights, all in one place. With experience spanning machine learning, data structuring, and intuitive frontend design, they’re not just writing code- they’re rethinking how evidence generation gets done.
What inspired the approach to building ANCHOR, and how does the team stay motivated by the work?
There’s a real openness to exploring advanced technologies like machine learning and AI to solve complex problems. The team thrives on experimenting with different ideas and approaches, which keeps the work creative and engaging. It’s more than coding, it’s about innovating in a space that’s traditionally been very manual and time-consuming.
How would you describe what ANCHOR does and the value it brings to users?
ANCHOR is a platform that streamlines the process of conducting a systematic literature review. It guides users through three main steps:
Criteria definition – users define their search strategy and eligibility criteria.
Screening – AI helps determine whether studies meet the criteria, starting with abstracts and then full texts.
Data extraction – key information is pulled from selected studies using AI tools, presented in a structured way for cross-study comparison.
By automating tasks like screening and extraction, ANCHOR cuts down on repetitive manual work and gives researchers more time to focus on analysis.
Systematic reviews can be time-consuming. How does automation in ANCHOR help speed things up without compromising control?
The step-by-step workflow keeps everything organised and in one place. Once search criteria are set, users can search databases, review hits, and pull publication data without leaving the platform, ANCHOR handles external communications in the background. A compliant search protocol can also be generated automatically.
For screening, users see each publication with relevant data and AI recommendations, while retaining full control to accept, reject, or compare decisions. Full-text PDFs can be retrieved or uploaded manually.
In the extraction phase, machine learning tools pull key data like baseline characteristics and outcomes. Everything is editable, so users can fine-tune as needed. The goal is to remove the manual grunt work and make space for more meaningful insights.
What technical challenges come with extracting structured data from PDFs, and how are those being solved?
Publications vary a lot in how they present data, tables, figures, scattered text. That inconsistency makes automation tricky. To address it, the team developed a pipeline combining machine learning and AI for detecting and extracting data, followed by a post-processing step that cleans and structures everything into a usable format.
Because AI models can be black boxes, transparency is a big priority. Multiple sources are used to cross-validate outputs, and a feature is being developed that will let users trace extracted data back to where it appears in the PDF. That way, users can trust what they’re seeing and know when manual review might be needed.
Why is mapping extracted data to standardised terms so critical when comparing study results?
Terms vary widely across publications, different spellings, abbreviations, or phrasing. Without standardisation, you can’t reliably compare or aggregate data. Mapping ensures that all versions of a concept are grouped together correctly.
But we are not just grouping, we are also preserving meaning. For instance, "current smoker" and "former smoker" shouldn’t be lumped into one "smoker" category if that distinction matters clinically. The team puts a lot of effort into getting that balance right so no important detail is lost in translation.
From a design perspective, how does ANCHOR aim to be intuitive and helpful for users?
With strong input from the evidence synthesis team; especially Maria Rizzo, the platform is designed around the actual workflows researchers follow. That understanding shapes how the interface is built: clean layouts, autosave, visual cues, and features like progress tracking.
Users always stay in control and can override AI suggestions at any point. Collaboration is also a key focus, with team features being developed to support joint reviews and shared projects.
What part of the development has been the most rewarding so far?
Building a workflow that turns messy, unstructured PDFs into clean, structured data has been both the hardest and most satisfying challenge. Seeing that technical backbone come to life through a user-friendly interface makes all the effort worthwhile and hopefully, that’s something users will feel the benefit of too.
Thank you for all your hard work team. We can’t wait to see the latest developments within ANCHOR.
Thank you!
If you have any questions about ANCHOR, please contact our Head of Evidence Synthesis, Maria Rizzo at maria.rizzo@gipam-health.com