TalentMap
Actionable talent insights from candidate-generated text, based on thousands of real profiles
Feature Highlights
Analysis view: Discover a candidate’s profile at a glance
Search module: Find the top 10 most similar profiles.
Mapping feature: See where a candidate stands across five key soft skills.
What this project illustrates
-
Turning psychological and NLP research into a directly usable talent tool
-
Designing models focused on interpretability and decision support
-
Balancing scientific validity with real-world recruitment constraints
-
Bridging textual signals, personality inference and talent decision-making
Context & intent
Context
Assessing soft skills is often subjective, time-consuming, and heavily dependent on human intuition.
In practice, what we commonly refer to as soft skills largely overlaps with personality traits—such as leadership, reliability, cooperation, or stress management.
Recruiters and managers rely on written or spoken candidate content to infer these traits, yet systematically extracting consistent personality signals from text remains difficult at scale.
Approach
TalentMap analyzes candidate-generated text using NLP and representation learning to infer personality traits and translate them into interpretable soft skills.
It also leverages similarity modeling to position a given profile within a broader talent space, enabling comparison, clustering, and mapping across candidates.
Value
-
Infer personality traits and soft skills directly from candidate text
-
Identify the most similar profiles from a CV or candidate database
-
Map a target profile against all others to reveal positioning and affinities
-
Reduce reliance on purely subjective human judgment
-
Support faster, more consistent talent decisions
Project management & key decisions
-
Choosing text-based inference to remain close to real hiring signals
-
Favoring interpretable trait-to-skill mappings over opaque scores
-
Using similarity search instead of rigid classification to reflect real-world hiring logic
-
Designing outputs to mirror how a colleague or recruiter would naturally compare profiles
-
Cost-controlled inference strategy: ~3 min first run, ~5 sec subsequent runs; 30 predictions/day cap
Tech stack
-
Frontend: Vite, Netlify
-
Backend: FastAPI, Python
-
NLP / ML: Transformer-based embeddings, XGB
-
Infra: Docker, Artifact Registry, GCS
Scientific background
TalentMap builds on research in personality psychology, psycholinguistics and natural language processing, showing that stable personality traits and soft skills can be inferred from linguistic patterns in spontaneous text.
Model performance on personality trait prediction for TalentMap
(Pearson correlation coefficients):
-
Openness · Adaptability: r = 0.16
-
Reliability · Organization (Conscientiousness): r = 0.17
-
Leadership · Influence (Extraversion): r = 0.25
-
Cooperation · Teamwork (Agreeableness): r = 0.14
-
Resilience · Stress Management (Neuroticism): r = 0.24
Typical self–other agreement reported in the literature (Pearson correlations):
-
Coworkers: r ≈ 0.25–0.30
-
Friends: r ≈ 0.40–0.45
-
Family members: r ≈ 0.45–0.50
-
Romantic partners / spouses: r ≈ 0.55–0.60
Youyou, W., Kosinski, M., & Stillwell, D. (2015). Computer-based personality judgments are more accurate than those made by humans. Proceedings of the National Academy of Sciences, 112(4), 1036-1040.