When you upload your CV to ProjexMaster, something happens that you will not find on any traditional job board: within 30 seconds, the platform analyzes your entire professional history and matches it against every active mandate using semantic AI — not keywords. The result is a precise fit score that tells both you and the hiring company exactly how well you align. Here is how it works, step by step.
The Matching Pipeline: Four Stages in 30 Seconds
ProjexMaster's matching engine runs a four-stage pipeline every time a new CV is uploaded or a new mandate is published. Each stage builds on the previous one, creating a multi-layered understanding of both talent and opportunity.
Stage 1: CV Upload and Text Extraction
When you upload your CV (PDF, DOCX, or plain text), ProjexMaster's document processing engine extracts structured information: your name, roles held, companies, dates, technologies, certifications, and — critically — the context in which you used each skill. This is not simple OCR. The system uses layout-aware parsing to understand tables, bullet points, and multi-column formats that trip up basic text extractors.
The extracted data is normalized into a structured profile: role history with durations, a skill taxonomy mapping, certification records, and a free-text summary of your professional narrative.
Stage 2: Semantic Embedding Generation
Here is where ProjexMaster diverges from every keyword-based platform. Your structured profile is passed through a large language model that generates a high-dimensional semantic embedding — a numerical representation of your professional identity in a 1536-dimensional vector space.
This embedding captures meaning, not just words. A CV that says "led agile transformation for a 200-person engineering organization" and one that says "drove organizational change management across multiple scrum teams" produce similar embeddings — because they describe similar experience, even though they share almost no keywords.
The same process runs on mandate briefings. Every job description is embedded into the same vector space, creating a shared language between talent and opportunity.
Stage 3: pgvector Similarity Search
With both talent and mandate embeddings stored in PostgreSQL via the pgvector extension, ProjexMaster performs approximate nearest-neighbor (ANN) searches at scale. When a new mandate is published, the system queries the vector index to find the top candidates whose embeddings are closest in semantic space.
This is blazingly fast. Even with tens of thousands of profiles in the database, the vector similarity search completes in under 50 milliseconds using HNSW (Hierarchical Navigable Small World) indexing. The result is a ranked shortlist of semantically relevant candidates — before any traditional filtering has even begun.
Stage 4: Multi-Dimensional Fit Score
The final stage combines semantic similarity with structured data to produce the composite fit score you see on every match. This score is not a single number from a black box — it is a weighted combination of four transparent components.
The Four Score Components
Every fit score on ProjexMaster is built from four independently calculated dimensions. Both freelancers and companies can see the breakdown, ensuring complete transparency.
Skill Match
40%Compares required skills and certifications against your verified profile. Accounts for skill level (basic, proficient, expert), recency of use, and years of experience per skill. A mandate requiring "SAFe SPC" and "Jira" will heavily weight candidates who hold those credentials and have used them recently.
Rate Alignment
25%Measures how well your rate expectations align with the mandate's budget. A perfect score means your rate falls within the client's range. The score degrades gracefully: a 10% mismatch might score 80%, while a 30%+ gap drops below 50%. This prevents wasted time on both sides.
Availability
20%Evaluates timing alignment between your availability window and the mandate's start date. Immediate availability for an urgent mandate scores 100%. A six-week gap might score 60%. This component also factors in location compatibility (onsite, hybrid, remote) and travel willingness.
Semantic Vector
15%The cosine similarity between your profile embedding and the mandate embedding. This captures the intangible "vibe" — industry context, communication style, project complexity level, and domain expertise that structured fields cannot fully represent. It is the component that catches matches keyword systems miss entirely.
The final score is a weighted sum: (Skill × 0.40) + (Rate × 0.25) + (Availability × 0.20) + (Vector × 0.15) = Fit Score. A score of 94% means exceptional alignment across all four dimensions.
Why Semantic Matching Beats Keyword Matching
Traditional job platforms rely on keyword matching: if the mandate says "Scrum Master" and your CV says "Scrum Master," you are a match. This approach has fundamental limitations that semantic matching overcomes.
The Synonym Problem
A mandate asking for an "Agile Coach" and a CV listing "Scrum Master and transformation lead" describe overlapping roles — but keyword matching sees zero overlap. Semantic matching understands they occupy similar positions in the professional landscape and scores them accordingly.
The Context Problem
Keywords cannot distinguish between "managed a team of 5" and "managed a program of 500." Both contain the word "managed," but they describe vastly different levels of responsibility. Semantic embeddings capture this context because the language model understands scale, complexity, and organizational hierarchy.
The Implicit Skill Problem
If your CV describes leading a banking IT transformation, a semantic model infers that you likely have experience with regulatory compliance, stakeholder management at C-level, vendor coordination, and risk management — even if none of those terms appear explicitly in your CV. Keyword matching would miss all of these implicit qualifications.
Transparency and Trust
We believe AI matching only works when both sides trust the results. That is why every fit score on ProjexMaster comes with a full breakdown: you can see exactly which skills contributed, where rate alignment stands, and how the semantic component scored. There are no hidden weights, no black-box overrides, and no pay-to-rank mechanisms.
Companies see the same transparency. When they receive a shortlist of candidates, each profile includes the four-component breakdown, allowing hiring managers to make informed decisions about which dimension matters most for their specific mandate.
The Result: Better Matches, Faster Placements
The combination of semantic embeddings, structured scoring, and transparent breakdowns produces measurably better outcomes. ProjexMaster matches reduce average time-to-shortlist from 5 days to 30 seconds, and first-interview acceptance rates exceed 85% — compared to an industry average of roughly 40% on keyword-based platforms.
For freelancers, this means fewer irrelevant mandate suggestions and more time spent on opportunities that genuinely fit. For companies, it means a shortlist they can trust from day one — no more sifting through 50 CVs to find 3 viable candidates.
Experience AI matching yourself
Upload your CV and see your fit score against active mandates — free for freelancers.
Create Your Free ProfileBOTFORCE Team
Insights on IT project management, freelance strategy, and the DACH tech market.