A psychology-grounded system that infers interpretable personality traits and soft skills from candidate-written content, then maps affinities across a talent pool.
RESEARCH→MODEL→PRODUCT
01 — THE CHALLENGE
A useful model starts with the right question.
CV keyword matching says little about how a person thinks or works. The objective was to structure richer talent signals while keeping the reasoning interpretable for human decision-makers.
Project anatomy
From candidate text to a psychometric talent map.
The system turns free text into a Big Five profile, positions it against a reference distribution and retrieves similar profiles from the talent pool.
Core pipeline
Select a step to understand how the data is transformed.
Selected step
The user enters or imports candidate content, then chooses the comparison depth and dimensions to explore.
Implemented features
Visible capabilities are connected to the technical components that make them possible.
01
Free-text analysis
The pipeline accepts candidate content without requiring an additional psychometric questionnaire.
02
Big Five profile
Five dimensions are returned as comparable scores and percentiles.
03
Reference comparison
Two texts can be analyzed together to display trait-level differences.
04
Top similar profiles
Neighbor count and quality mode can be configured in the interface.
05
Adaptive map
The projection changes with the number of selected traits and keeps reproducible coordinates.
06
Complementary visuals
Big Five radar, results list, differences and interactive map support the decision.
Stack by responsibility
Each technology is placed within its role in the system.
01
React · Vite
Uploads, settings, maps and tabbed navigation.
02
Recharts · Plotly
Big Five profile and interactive talent-pool projection.
03
FastAPI · Pydantic
Prediction, comparison, search and mapping endpoints.
04
Sentence Transformers
Normalized multilingual MiniLM embeddings.
05
XGBoost · Scikit-learn
Trait-level models, standardization and similarity computation.
06
UMAP · GCS
2D projection and optional production artifact loading.
NLP and HR decision supportBased on the current codebase · Simplified representation
From technology to use
Follow the journey from text to profile and map.
The experience stages candidate-text analysis, the five personality dimensions and the profile’s position among comparable talents.
Educational demonstration: data is local and some computations are simulated or accelerated.
01Interactive demonstration — analyse and map a profile
04 — OUTCOME AND IMPACT
“A decision-support workflow that turns unstructured candidate language into comparable insights — without replacing human judgment.”