1. A computer-implemented method executed by one or more hardware processors for multi-role-based deterministic automated resume classification, the method comprising: receiving a plurality of job description documents corresponding to multiple organisational roles through a user interface module (100); converting each job description document into machine-readable text and extracting a role identifier and one or more role scope definitions through a job description parsing module (110); mapping the extracted role identifier to a predefined organisational role taxonomy stored in a role taxonomy repository (130) to obtain a valid taxonomy role identifier; receiving a plurality of candidate resumes in electronic document formats through the user interface module (100); extracting resume text and candidate attributes from each resume by a resume extraction module (120), constructing a combined multi-role evaluation input comprising the role scope definitions corresponding to the valid taxonomy role identifiers and optionally recruiter-defined runtime instructions through a prompt construction module (140); processing the combined multi-role evaluation input using a machine learning model to generate, for each resume, at least a compatibility score and an inferred role association through an artificial intelligence evaluation module (150); normalising the inferred role association to the predefined organisational role taxonomy using deterministic and similarity-based mapping to obtain a normalised role identifier by a role normalisation module (160); retrieving predetermined threshold parameters from memory and applying deterministic threshold-based decision rules to the compatibility score to classify the resume into one of predefined outcome categories comprising Shortlisted, Waiting, and Rejected by a deterministic decision engine (170); and generating and storing in a database a structured evaluation record including at least the compatibility score, the normalised role identifier, and the classification category by a reporting and audit module (190).

2. The method as claimed in claim 1, wherein extracting the role identifier comprises identifying a primary role designation from a header portion of the job description document or from a file name associated with the job description document and cleaning the designation for standardisation.

3. The method as claimed in claim 1, wherein mapping the extracted role identifier to the predefined organisational role taxonomy comprises performing exact string matching, case-insensitive comparison and similarity-based matching including edit-distance computation.

4. The method as claimed in claim 1, wherein constructing the combined multi-role evaluation input comprises consolidating a plurality of role scope definitions into a single evaluation context to enable comparative evaluation in a single inference execution.

5. The method as claimed in claim 4, wherein the single inference execution reduces the number of inference calls required per resume when compared to role-wise separate evaluation.

6. The method as claimed in claim 1, wherein the recruiter-defined runtime instructions are captured through a user interface and programmatically injected into a predefined instruction segment of an evaluation template without retraining the machine learning model.

7. The method as claimed in claim 1, wherein applying deterministic threshold-based decision rules comprises comparing the compatibility score with a first predetermined threshold and a second predetermined threshold stored in configuration memory.

8. The method as claimed in claim 1, further comprising automatically creating category-specific storage directories and moving or copying resume files into the directories based on the classification category.

9. A computer system comprising one or more processors and a memory storing executable instructions and when executed by the one or more processors, cause the system to perform multi-role-based deterministic automated resume classification, the system comprising: a job description parsing module (110) configured to extract role identifiers and role scope definitions and map the role identifiers to a predefined organisational role taxonomy; a resume extraction module (120) configured to extract resume text and candidate attributes from electronic resume documents; a prompt construction module (140) configured to construct a combined multi-role evaluation input; an artificial intelligence evaluation module (150) configured to generate a compatibility score and an inferred role association for each resume; a role normalisation module (160) configured to map the inferred role association to a normalised role identifier in the predefined organisational role taxonomy; and a deterministic decision engine (170) configured to apply predetermined threshold-based decision rules to classify each resume into predefined outcome categories.

10. The system as claimed in claim 9, further comprising a document segregation module (180) configured to automatically move or copy resume files into category-specific storage locations based on the classification category.

11. The system as claimed in claim 9, wherein the system is configured to store structured evaluation records including compatibility score, normalised role identifier, classification category, and batch identifier in a database.

12. A non-transitory computer-readable medium storing computer-executable instructions which, when executed by one or more hardware processors, cause the one or more hardware processors to perform the method as claimed in any one of claims 1 to 8.

Description:METHOD AND SYSTEM FOR MULTI‑ROLE BASED DETERMINISTIC AUTOMATED RESUME CLASSIFICATION

TECHNICAL FIELD

[0001] The present invention relates to computer-implemented recruitment systems and methods for automated resume screening. More particularly, the invention relates to multi-role resume classification using a combination of machine learning-based evaluation and deterministic rule-based decision logic.

BACKGROUND

[0002] Recruitment processes in medium and large organisations typically involve screening a high volume of candidate resumes against one or more job descriptions. In conventional practice, recruiters manually review resumes, analyse educational background and professional experience, compare the same with defined job requirements and determine whether a candidate should be shortlisted, kept in consideration or rejected. Such manual evaluation is time-consuming and leads to increased hiring cycle duration. Further, manual screening is dependent on individual interpretation and may result in inconsistency across different recruiters.

[0003] Organisations frequently conduct recruitment drives for multiple roles simultaneously. A plurality of job descriptions may correspond to different departments, technology stacks, functional domains, or seniority levels. In such multi-role scenarios, a resume may be relevant to more than one role. Conventional practice requires evaluating the same resume separately against each role, which increases processing time and computational overhead when automated systems are used.

[0004] Conventional automated resume screening tools are known in the art and such tools mostly rely on keyword-based matching techniques. These systems typically identify occurrences of predefined keywords in a resume and compare them with keywords present in a job description. However, keyword-based approaches may fail to capture contextual meaning, semantic relevance, experience depth or domain alignment. As a result, suitable candidates may be overlooked due to vocabulary mismatch or unsuitable candidates may be selected due to superficial keyword repetition.

[0005] The patent document US11308142B2 discloses a system and method for automated resume evaluation and candidate ranking using keyword extraction and weighted scoring techniques. The disclosed system compares resume content with predefined job requirements and generates a suitability score. However, the evaluation is performed on a per-role basis and relies primarily on keyword frequency and rule-based weighting.

[0006] The patent document US12141757 describes a recruitment analytics platform employing machine learning models for candidate-job matching and predictive hiring recommendations. The system generates match scores based on historical hiring data and skill alignment analysis. The document discloses machine learning-based scoring, but it does not teach deterministic threshold-based enforcement applied after semantic evaluation to produce predefined outcome categories.

[0007] Recently developed solutions employ artificial intelligence models for semantic evaluation of resumes. Such systems may generate compatibility scores or role suggestions, but the evaluation process may operate as a non-transparent model-driven output without deterministic enforcement of decision boundaries. In practical enterprise environments, organisations require reproducible and auditable classification logic. If classification decisions vary without controlled threshold enforcement or structured role normalisation, downstream workflow integration becomes difficult.

[0008] Further, known AI-based systems typically evaluate a resume against a single job description per inference cycle. Where multiple roles are involved, separate inference executions are performed for each role. Such one-to-one evaluation approach is computationally inefficient for multi-role recruitment scenarios and increases processor utilisation, memory consumption, and network overhead in distributed or cloud-based computing environments.

[0009] Organisations generally maintain a predefined role taxonomy comprising approved role identifiers and standardised role names. Existing AI-driven systems may output free-text role suggestions, abbreviations or semantically similar variations that do not match the organisation’s approved taxonomy. This creates inconsistencies in reporting, workflow integration and database storage.

[0010] Recruiters may also require runtime flexibility to prioritise certain criteria, such as educational specialisation, domain exposure, certification requirements or years of experience. In many existing systems, modifying such priorities requires retraining of the model or technical reconfiguration, resulting in operational delay.

[0011] In addition to classification, recruitment workflows require operational outputs such as structured reports, systematic segregation of resume documents, storage of batch metadata, and reproducible evaluation records. Conventional approaches often require manual post-processing, including manual folder creation and spreadsheet preparation, leading to inefficiency and potential error.

[0012] To address the above limitations, there is a need for a computer-implemented system and method that:
• performs comparative evaluation of a resume across multiple predefined roles in a single inference execution;
• normalises role outputs to a controlled organisational role taxonomy;
• enforces deterministic threshold-based decision boundaries;
• reduces computational overhead associated with repeated inference calls;
• generates structured evaluation records with stored taxonomy version and threshold parameters; and
• automatically performs file system operations and report generation for recruitment workflow integration.

SUMMARY

[0013] The present invention provides a computer-implemented method and system for multi-role-based deterministic automated resume classification. The invention is particularly directed to large-scale recruitment scenarios involving a plurality of job descriptions and a plurality of candidate resumes.

[0014] In one embodiment, the invention receives multiple job description documents, extracts role identifiers and role scope definitions, and maps the same to a predefined organisational role taxonomy stored as structured data in a database or configuration repository.

[0015] Candidate resumes in electronic formats are converted into machine-readable text and candidate attributes including education, experience, skills, and domain indicators are extracted using text parsing techniques.

[0016] A prompt construction module consolidates the plurality of role scope definitions into a single evaluation context. Optionally, recruiter-defined runtime instructions may be captured and incorporated into the evaluation input without retraining the artificial intelligence model.

[0017] The combined evaluation input is processed using a machine learning model comprising a transformer-based language model architecture configured for contextual reasoning. The model generates at least a compatibility score and an inferred role association for each resume.

[0018] The inferred role association is processed by a role normalisation module that performs deterministic string comparison and similarity-based mapping, including token overlap scoring and edit-distance-based similarity computation, to map the inferred role to the closest valid entry in the predefined role taxonomy.

[0019] A deterministic decision engine applies hard boundary conditions using a first predetermined threshold and a second predetermined threshold stored in memory or a database. Based on comparison of the compatibility score with the predetermined thresholds, each resume is classified into predefined outcome categories comprising Shortlisted, Waiting or Rejected.

[0020] In one embodiment, simultaneous multi-role evaluation reduces the number of inference executions required per resume from multiple role-wise executions to a single execution. This reduces processor utilisation, memory consumption, and network overhead in distributed computing environments.

[0021] The system further generates structured evaluation records, stores taxonomy version identifiers and threshold parameter values used for a batch, and automatically performs file system operations to segregate resume documents into category-specific storage locations. Tabular reports and downloadable archives may be generated programmatically.

BRIEF DESCRIPTION OF THE DRAWINGS

[0022] FIG. 1 is a schematic block diagram illustrating the overall system architecture.
[0023] FIG. 2 is a flow diagram illustrating an end-to-end method of processing.
[0024] FIG. 3 is a schematic diagram illustrating the role taxonomy creation and role normalisation process.
[0025] FIG. 4 is a workflow diagram illustrating automated post-classification actions.

DETAILED DESCRIPTION

System Architecture

[0026] The present invention will now be described in detail with reference to the accompanying drawings. The embodiments described herein are illustrative and enable a person skilled in the art to implement the invention without undue experimentation. The invention is implemented in a computing environment comprising one or more hardware processors, memory units, persistent storage devices, communication interfaces and input-output interfaces. The memory stores executable instructions which, when executed by the one or more processors, cause the system to perform the steps described herein. The invention may be deployed in an on-premise server infrastructure, private cloud, public cloud or hybrid computing environment.

[0027] FIG. 1 illustrates a schematic block diagram of the system architecture for multi-role-based deterministic automated resume classification. As shown in FIG. 1, the system comprises a user interface module (100), a job description parsing module (110), a resume extraction module (120), a role taxonomy repository (130), a prompt construction module (140), an artificial intelligence evaluation module (150), a role normalisation module (160), a deterministic decision engine (170), a document segregation module (180), and a reporting and audit module (190). The modules may be implemented as software components executed by the processors and may communicate via application programming interfaces (APIs), structured data exchange formats, or internal service calls.

[0028] The user interface module (100) enables authorised recruitment personnel to initiate screening batches. In one embodiment, the user interface is implemented as a web-based application accessible over a secure network. The user interface allows uploading of multiple job description documents and multiple candidate resumes in electronic formats. The module further provides an optional input field for capturing recruiter-defined runtime instructions. These instructions may specify prioritisation of certain qualifications, domain expertise, certifications, or minimum experience requirements for a particular screening batch. Upon initiation of a batch, a unique batch identifier is generated and stored in system memory for tracking and audit purposes.

Job Description Parsing Module

[0029] The job description parsing module (110) receives job description documents in formats such as PDF, DOC, DOCX, or text files and converts them into machine-readable text using document parsing engines. The module identifies structured sections including role title, educational qualifications, experience requirements, responsibilities, and domain-specific keywords. In one embodiment, the module extracts a primary role designation from header sections or labelled text segments such as “Role”, “Position”, or “Job Title”.

[0030] The extracted role string is processed to remove extraneous characters, location indicators, and formatting inconsistencies. The cleaned role string is compared against entries in the role taxonomy repository (130). The role taxonomy repository stores a predefined organisational role taxonomy comprising structured entries including role identifier, standardised role name, optional role description, and taxonomy version identifier. The repository may be implemented using a relational database or structured configuration storage.

[0031] Mapping between the extracted role string and taxonomy entries is performed using deterministic and similarity-based matching. The system first attempts exact and case-insensitive string matching. If no match is found, token overlap scoring and edit-distance similarity computation such as Levenshtein distance may be applied. In certain embodiments, cosine similarity over tokenised vector representations may also be used. The closest valid taxonomy entry is selected, and the corresponding role identifier is assigned to the job description.

[0032] FIG. 3 illustrates the role taxonomy creation and normalisation mechanism. The figure depicts storage of structured taxonomy entries and mapping of both job description role strings and AI-inferred role outputs to predefined taxonomy identifiers. Taxonomy version information is associated with each screening batch to ensure reproducibility of classification outcomes.

Resume Extraction Module

[0033] The resume extraction module (120) receives candidate resumes in electronic formats and converts them into machine-readable text. The module identifies candidate name, contact information, educational qualifications, work experience details, skills, certifications, and project descriptions using pattern recognition, section-based parsing, and natural language processing techniques. The extracted information is stored in structured or semi-structured form in database storage.

[0034] The structured storage of resume attributes enables deterministic rule evaluation at later stages and supports reporting and audit functionality. The extraction process is executed programmatically using server-side processing and produces structured data objects associated with each resume file.

Multi-Role Prompt Construction Module

[0035] The prompt construction module (140) consolidates multiple role scope definitions derived from job descriptions into a single evaluation context. Unlike conventional approaches that perform separate inference calls per role, the present invention constructs a unified evaluation input that includes all relevant role descriptions.

[0036] Optional recruiter-defined runtime instructions are programmatically injected into a predetermined instruction section of the evaluation template. This injection is performed dynamically at runtime without retraining the artificial intelligence model. The injected instructions are stored as part of batch metadata.

[0037] By consolidating multiple roles into a single evaluation input, the system reduces the number of inference executions required per resume from N separate executions to a single execution, where N represents the number of roles. This reduction decreases processor utilisation, memory allocation overhead, and network communication load in distributed computing environments. The reduction in inference calls constitutes a measurable technical improvement in computational efficiency.

[0038] FIG. 2 illustrates the end-to-end method flow. The figure depicts receipt of job descriptions and resumes, extraction of role identifiers and resume attributes, construction of the combined evaluation input, execution of a single inference cycle, role normalisation, deterministic threshold application, and generation of structured outputs. Data flow arrows illustrate structured data transfer between modules.

Artificial Intelligence Evaluation Module

[0039] The artificial intelligence evaluation module (150) comprises a machine learning model implemented using a transformer-based language model architecture. The model operates in inference mode and receives the consolidated evaluation input comprising role scope definitions, resume text, and optional runtime instructions.

[0040] The model performs contextual semantic reasoning and generates structured output including a compatibility score within a predefined numeric range, an inferred role association selected from the provided role list, and a justification summary describing the basis of evaluation. The structured output is formatted in machine-readable format such as JSON to enable deterministic downstream processing.

Role Normalisation Module

[0041] The role normalisation module (160) ensures strict alignment of inferred role outputs with the predefined organisational role taxonomy. The module applies exact string matching, case-insensitive comparison, token overlap scoring, edit-distance similarity computation, and optionally cosine similarity over vectorised tokens.

[0042] The closest valid taxonomy entry is selected as the final normalised role identifier. This deterministic mapping prevents generation of uncontrolled or non-standard role names and ensures consistency in database storage and reporting.

Deterministic Threshold-Based Classification Module

[0043] The deterministic decision engine (170) retrieves predetermined threshold parameters from memory or configuration storage. In one embodiment, a first predetermined threshold and a second predetermined threshold are defined.

[0044] The compatibility score is compared with these thresholds. If the score is greater than or equal to the first predetermined threshold, the resume is classified as Shortlisted. If the score lies between the first and second thresholds, the resume is classified as Waiting. If the score is below the second threshold, the resume is classified as Rejected.

[0045] Additional deterministic rules may be applied based on structured resume attributes. For example, a minimum educational qualification or minimum years of experience may be required for Shortlisted classification. These rule parameters are stored in configuration storage and executed programmatically.

Automated Document Segregation and Reporting Module

[0046] Upon final classification, the document segregation module (180) automatically performs file system operations. The module checks for existence of category-specific directories and creates them if absent. Resume files are programmatically moved or copied into corresponding directories using operating system-level APIs. These operations produce a tangible technical effect in storage organisation.

[0047] The reporting and audit module (190) generates structured tabular reports in formats such as Excel or CSV. The report includes candidate identifier, normalised role identifier, compatibility score, classification category, justification summary, batch identifier, taxonomy version identifier, threshold parameter values, and processing timestamp.

[0048] FIG. 4 illustrates the post-classification workflow including structured database storage, report generation, and optional creation of downloadable archive files for each classification category. The figure demonstrates integration of file segregation, metadata storage, and reporting.

Structured Record Storage Module

[0049] The system stores structured evaluation records in a database. Each record includes compatibility score, inferred role, normalised role identifier, classification category, justification summary, recruiter-defined instructions, taxonomy version identifier, threshold parameter values, batch identifier, and timestamp. This structured storage ensures that screening outcomes are reproducible and auditable.

Example 1: High-Volume Multi-Role Campus Recruitment

[0050] In one embodiment, an organisation conducts a campus recruitment drive involving roles such as Software Engineer, Data Analyst, and DevOps Engineer. Hundreds of resumes are uploaded in a single batch. The system extracts role identifiers from job descriptions and constructs a combined multi-role evaluation input. Each resume is evaluated through a single inference execution, reducing computational overhead. Compatibility scores are generated and deterministic thresholds are applied. Resumes are automatically segregated into Shortlisted, Waiting, and Rejected folders, and a structured report is generated for interviewer allocation.

Example 2: Enterprise Recruitment with Dynamic Prioritisation

[0051] In another embodiment, an enterprise screens resumes for a cloud engineering role. For a specific batch, the recruiter prioritises cloud certifications and distributed systems experience. The runtime instruction is injected into the evaluation template without retraining the model. The system performs semantic evaluation, applies deterministic thresholds, and enforces minimum certification constraints. Structured reports and segregated folders are generated for managerial review.

Example 3: Integration with Applicant Tracking System

[0052] In a further embodiment, resumes received through an applicant tracking system are periodically exported and processed in hourly batches. The system evaluates each resume using single-cycle multi-role reasoning, assigns normalised role identifiers, applies deterministic classification, and stores evaluation records. The final classification and role mapping are transmitted back to the applicant tracking system via API, enabling automated workflow progression.

[0053] The invention provides measurable technical effects including reduction in inference executions, reduced processor and memory utilisation, deterministic enforcement of classification logic at server level, structured database storage with taxonomy and threshold version control, and automated file system manipulation. The integration of semantic reasoning with deterministic rule enforcement produces a controlled and reproducible technical workflow.

[0054] It will be understood that modifications and variations may be made without departing from the scope of the invention. The described embodiments enable a person skilled in the art to implement the invention using standard computing components, text extraction libraries, similarity computation techniques, and machine learning inference engines.
, C , C , Claims:We claim:
1. A computer-implemented method executed by one or more hardware processors for multi-role-based deterministic automated resume classification, the method comprising:
receiving a plurality of job description documents corresponding to multiple organisational roles through a user interface module (100);
converting each job description document into machine-readable text and extracting a role identifier and one or more role scope definitions through a job description parsing module (110);
mapping the extracted role identifier to a predefined organisational role taxonomy stored in a role taxonomy repository (130) to obtain a valid taxonomy role identifier;
receiving a plurality of candidate resumes in electronic document formats through the user interface module (100);
extracting resume text and candidate attributes from each resume by a resume extraction module (120),
constructing a combined multi-role evaluation input comprising the role scope definitions corresponding to the valid taxonomy role identifiers and optionally recruiter-defined runtime instructions through a prompt construction module (140);
processing the combined multi-role evaluation input using a machine learning model to generate, for each resume, at least a compatibility score and an inferred role association through an artificial intelligence evaluation module (150);
normalising the inferred role association to the predefined organisational role taxonomy using deterministic and similarity-based mapping to obtain a normalised role identifier by a role normalisation module (160);
retrieving predetermined threshold parameters from memory and applying deterministic threshold-based decision rules to the compatibility score to classify the resume into one of predefined outcome categories comprising Shortlisted, Waiting, and Rejected by a deterministic decision engine (170); and
generating and storing in a database a structured evaluation record including at least the compatibility score, the normalised role identifier, and the classification category by a reporting and audit module (190).
2. The method as claimed in claim 1, wherein extracting the role identifier comprises identifying a primary role designation from a header portion of the job description document or from a file name associated with the job description document and cleaning the designation for standardisation.
3. The method as claimed in claim 1, wherein mapping the extracted role identifier to the predefined organisational role taxonomy comprises performing exact string matching, case-insensitive comparison and similarity-based matching including edit-distance computation.
4. The method as claimed in claim 1, wherein constructing the combined multi-role evaluation input comprises consolidating a plurality of role scope definitions into a single evaluation context to enable comparative evaluation in a single inference execution.
5. The method as claimed in claim 4, wherein the single inference execution reduces the number of inference calls required per resume when compared to role-wise separate evaluation.
6. The method as claimed in claim 1, wherein the recruiter-defined runtime instructions are captured through a user interface and programmatically injected into a predefined instruction segment of an evaluation template without retraining the machine learning model.
7. The method as claimed in claim 1, wherein applying deterministic threshold-based decision rules comprises comparing the compatibility score with a first predetermined threshold and a second predetermined threshold stored in configuration memory.
8. The method as claimed in claim 1, further comprising automatically creating category-specific storage directories and moving or copying resume files into the directories based on the classification category.
9. A computer system comprising one or more processors and a memory storing executable instructions and when executed by the one or more processors, cause the system to perform multi-role-based deterministic automated resume classification, the system comprising:
a job description parsing module (110) configured to extract role identifiers and role scope definitions and map the role identifiers to a predefined organisational role taxonomy;
a resume extraction module (120) configured to extract resume text and candidate attributes from electronic resume documents;
a prompt construction module (140) configured to construct a combined multi-role evaluation input;
an artificial intelligence evaluation module (150) configured to generate a compatibility score and an inferred role association for each resume;
a role normalisation module (160) configured to map the inferred role association to a normalised role identifier in the predefined organisational role taxonomy; and
a deterministic decision engine (170) configured to apply predetermined threshold-based decision rules to classify each resume into predefined outcome categories.
10. The system as claimed in claim 9, further comprising a document segregation module (180) configured to automatically move or copy resume files into category-specific storage locations based on the classification category.
11. The system as claimed in claim 9, wherein the system is configured to store structured evaluation records including compatibility score, normalised role identifier, classification category, and batch identifier in a database.
12. A non-transitory computer-readable medium storing computer-executable instructions which, when executed by one or more hardware processors, cause the one or more hardware processors to perform the method as claimed in any one of claims 1 to 8.