UtilityMax / API / Job Spec Best Candidate

Job Spec Best Candidate · API Endpoint

AI Recruitment Tools, Configured by You.

Most AI recruitment tools give you the vendor's ranking. UtilityMax gives you yours. The framework decomposes candidate fit into eight calibrated components, and your organisation defines which ones gate the decision, which contribute additively, and at what weights — through a configurable API for transparent candidate scoring.

Request Trial Access → See the Demo ↓ Read the Methodology →

Published research framework · Bearer-token API · Currently in trial with selected partners

The Problem

Why generic AI recruitment tools and AI hiring software get the wrong candidate first

Most AI recruitment tools give you their ranking. The vendor decides what matters — usually some opaque blend of keyword match, seniority signal, and pattern similarity to past hires. The output is a single number per candidate. The reasoning is invisible. The weighting is fixed.

Real organisations do not hire that way. A research-stage biotech weights different things than a non-profit. A regulated firm weights different things than a creative agency. A team building out their platform weights different things than a team executing on a shipped product. The same candidate is the right hire for one and the wrong hire for the other. Generic AI screening cannot represent that.

UtilityMax is configurable. The framework defines eight calibrated dimensions every candidate is scored against. Your organisation defines which dimensions are non-negotiable, which contribute additively, and at what weight. Two organisations using the same five candidates against the same job will produce two different rankings. The framework respects the difference. That is the point.

The Framework

Eight calibrated components for configurable AI recruitment tools

Each call to the API decomposes candidate fit into eight binary chance nodes. Each node has an explicit, observable definition. The probability that the node holds for a given candidate is estimated by the model conditioned on the position specification and the candidate's profile.

Component Definition Depends on
Education The candidate holds the educational qualification (or stated equivalent) named in the position specification. Position spec, candidate profile
Eligibility The candidate satisfies every hard eligibility constraint stated in the specification (work authorisation, language proficiency, clearances, licensure). Position spec, candidate profile
Technical skills For every skill, tool, or methodology the spec names as required, the candidate's record contains direct evidence of prior hands-on application. Position spec, candidate profile
Domain The candidate's prior experience has been substantively conducted in the domain or subject area named in the specification. Position spec, candidate profile
Validated output The candidate's record contains validated deliverables of the type expected at the seniority level the specification implies. Position spec, candidate profile
Environmental pedigree The candidate's prior experience has been conducted under leadership and within an organisational environment with a sustained track record in adjacent fields. Position spec, candidate profile
Trajectory The candidate's stated motivations and prior choices are consistent with the goals stated in the specification. Position spec, candidate profile
Cultural alignment The candidate's record is consistent with the cultural and operational expectations stated by the hosting organisation. Position spec, organisation context, candidate profile

Each component takes one of five importance levels in your configuration: non_negotiable, very_high, high, moderate, or low. Components marked non-negotiable act as multiplicative gates — fail one and the candidate's overall score collapses. The remaining components contribute additively at fixed weights (very_high = 4, high = 2, moderate = 1, low = 0.5). Three components — validated output, environmental pedigree, and trajectory — cannot be set as non-negotiable, because they are matters of degree rather than yes/no disqualifiers.

O(a)  =  ∏i ∈ gates P(Xi = 1  |  A = a, K)  ×  ∑j ∈ contributors wj · P(Xj = 1  |  A = a, K)

The gated additive form is what makes the framework configurable. Multiplicative gates encode hard requirements; the weighted sum encodes the organisation's priorities among the rest. The same candidate scored against two different configurations produces two different scores — and across multiple candidates, two different rankings.

Applied Result

Live demonstration: same candidates, two organisations, two rankings

To demonstrate the configurable ranker, we ran the same five candidates against the same role under two different organisational contexts. The candidates do not change. The role does not change. What changes is the organisation hosting the role, and the configuration the organisation chose. The numbers below are real — returned by the production API.

Configuration A

Build the platform

A small, well-funded biotech currently scaling its technical platform after a recent funding round. The team is in build mode: the priority is delivering a clinically deployable pipeline that meets the technical specifications the company has committed to.

  • EducationGATE
  • EligibilityGATE
  • Technical skillsVERY HIGH  ×4
  • Validated outputVERY HIGH  ×4
  • DomainHIGH  ×2
  • Environmental pedigreeHIGH  ×2
  • TrajectoryMODERATE  ×1
  • Cultural alignmentLOW  ×0.5
Configuration B

Embody the mission

A non-profit research consortium developing the same detection capability for deployment in low- and middle-income healthcare systems. The mission is the entire reason for being. Affordability and global accessibility shape every technical decision.

  • EducationGATE
  • EligibilityGATE
  • Cultural alignmentVERY HIGH  ×4
  • Technical skillsHIGH  ×2
  • TrajectoryHIGH  ×2
  • DomainMODERATE  ×1
  • Validated outputLOW  ×0.5
  • Environmental pedigreeLOW  ×0.5
The five candidates
  1. 01 Anders — MSc Bioinformatics (Copenhagen). Plasma proteomics methodologist at the Mann lab; cardiovascular focus, not cancer.
  2. 02 Yuki — MSc Computational Biology (Tokyo). ctDNA dynamics in early breast cancer; genomics rather than proteomics.
  3. 03 Aisha — MSc Computational Biology (LUMS, Pakistan). End-to-end LC-MS pipeline for breast cancer biomarker discovery; open-sourced.
  4. 04 Eleni — MSc Molecular Biotechnology (Heidelberg). Plasma proteomics in breast cancer at EMBL Savitski group; two 2025 papers.
  5. 05 Léa — MSc Cancer Biology (ENS Lyon). Liquid biopsy panel for breast cancer at Centre Léon Bérard; clinical embedding.
Build the platform
Embody the mission
1
Eleni
Validated output: 0.96
11.70
2
Anders
Environmental pedigree: 0.98
10.95
3
Aisha
Domain: 0.97
10.62
4
Yuki
Validated output: 0.90
10.43
5
Léa
Trajectory: 0.88
8.70
1
Aisha
Cultural alignment: 0.95
8.50
2
Léa
Domain: 0.93
5.46
3
Eleni
Validated output: 0.96
5.33
4
Yuki
Domain: 0.92
5.24
5
Anders
Validated output: 0.92
4.80
Same candidates· Same job· Different organisations· Different #1

Both organisations gate on education and eligibility — every shortlisted candidate must clear those bars. Beyond that, the configurations diverge sharply. Build the platform weights validated output and technical skills at 4 each, rewarding Eleni's two 2025 papers and Anders' Mann-lab pedigree. Embody the mission weights cultural alignment at 4, rewarding Aisha's open-source LMIC pipeline and decisively penalising candidates whose records contain no signal of mission orientation. Anders, who tops the technical pedigree dimension, ranks last under Embody not because his work is poor — it is excellent — but because the configuration does not value what he is excellent at. The framework respects the difference.

Run a configuration like this against your own role and candidate pool. Trial access opens with a 30-minute onboarding call where the configuration is constructed together.
Request Trial Access →
View full Build the platform brief and configuration

Position specification

Doctoral Researcher Position: Plasma Proteomic Biomarkers for Early Cancer Detection

Four-year doctoral researcher position hosted by a small computational diagnostics company developing a plasma proteomic platform for early-stage cancer detection. The position is jointly supervised with a European academic partner that will award the doctoral degree.

Project scope. The successful candidate will contribute to the development of a clinically deployable biomarker pipeline based on plasma LC-MS proteomics. Work spans the full pipeline: sample handling protocols, data acquisition and processing, statistical biomarker discovery, machine learning for predictive classifier development, integration with clinical metadata, and validation against independent cohorts. The candidate will be expected to contribute to platform engineering decisions — assay parameters, processing infrastructure, classifier architecture — rather than working only at the statistical end of the pipeline.

Required qualifications. MSc in bioinformatics, computational biology, molecular biology, biomedical sciences, data science, or a closely related field. Awarded or expected by start date. Candidates already holding a PhD are not eligible.

Eligibility. Standard EU doctoral mobility rules apply: the candidate must not have resided or carried out their main activity in the host country for more than 12 months in the 36 months immediately before recruitment. Working language: English.

Required skills. Programming proficiency in Python and/or R. Solid grounding in applied statistics. Hands-on experience with at least one of: mass-spectrometry proteomics, machine learning, multi-omics data integration, clinical data analysis. Demonstrated ability to deliver an end-to-end research project (defended thesis, conference contribution, or equivalent).

Organisation context

The host is a small (under ten people), well-funded biotech currently scaling its technical platform following a recent funding round. The team is in build mode: the immediate priority is delivering a robust, clinically deployable pipeline that meets the technical specifications the company has committed to.

Operating values. The team operates with a strong delivery orientation. Successful contributors are technically deep, ship code that runs in production, and engage substantively with the methodological choices that determine platform performance. The company values verifiable technical capability — peer-reviewed publications, released code, reproducible benchmarks — over potential or aspiration.

What success looks like. Successful hires bring substantive technical depth that is verifiable from their record (publications, shipped artefacts, presented work), have demonstrated end-to-end delivery on comparable problems, and operate well in a small team where every hire is expected to take meaningful ownership of a piece of the platform.

Per-candidate component scores

Eleni — score 11.70 (rank 1)
ComponentP(X=1)
Education0.99
Eligibility0.95
Technical skills0.95
Domain0.88
Validated output0.96
Environmental pedigree0.97
Trajectory0.72
Cultural alignment0.75
Anders — score 10.95 (rank 2)
ComponentP(X=1)
Education0.99
Eligibility0.95
Technical skills0.92
Domain0.55
Validated output0.92
Environmental pedigree0.98
Trajectory0.78
Cultural alignment0.88
Aisha — score 10.62 (rank 3)
ComponentP(X=1)
Education0.99
Eligibility0.92
Technical skills0.96
Domain0.97
Validated output0.85
Environmental pedigree0.55
Trajectory0.92
Cultural alignment0.92
Yuki — score 10.43 (rank 4)
ComponentP(X=1)
Education0.99
Eligibility0.95
Technical skills0.78
Domain0.72
Validated output0.90
Environmental pedigree0.85
Trajectory0.82
Cultural alignment0.82
Léa — score 8.70 (rank 5)
ComponentP(X=1)
Education0.93
Eligibility0.95
Technical skills0.65
Domain0.78
Validated output0.72
Environmental pedigree0.80
Trajectory0.88
Cultural alignment0.65
View full Embody the mission brief and configuration

Position specification

Doctoral Researcher Position: Plasma Proteomic Biomarkers for Early Cancer Detection in Low-Resource Settings

Four-year doctoral researcher position hosted by a non-profit research consortium developing a plasma proteomic detection platform with the explicit aim of clinical deployment in low- and middle-income healthcare systems. The position is jointly supervised with an academic partner that will award the doctoral degree.

Project scope. The successful candidate will contribute to a biomarker pipeline whose technical decisions are constrained by deployability in resource-limited settings: assay choices, sample preparation, processing infrastructure, and classifier complexity are all evaluated against eventual clinical use in environments without specialist mass-spectrometry facilities or high-bandwidth data infrastructure.

Required qualifications. MSc in bioinformatics, computational biology, molecular biology, biomedical sciences, data science, or a closely related field. Awarded or expected by start date. Candidates already holding a PhD are not eligible.

Eligibility. Standard EU doctoral mobility rules apply: the candidate must not have resided or carried out their main activity in the host country for more than 12 months in the 36 months immediately before recruitment. Working language: English.

Required skills. Programming proficiency in Python and/or R. Solid grounding in applied statistics. Hands-on experience with at least one of: mass-spectrometry proteomics, machine learning, multi-omics data integration, clinical data analysis. Demonstrated ability to deliver an end-to-end research project (defended thesis, conference contribution, or equivalent).

Desirable. Direct experience with LC-MS proteomics data analysis. Prior work in oncology or circulating-biomarker domains. Demonstrated orientation toward globally accessible deployment — work conducted in or with low- and middle-income institutions, designs for resource-constrained settings, open-source release of research artefacts.

Organisation context

The host is a small non-profit research consortium whose entire reason for being is the delivery of early cancer detection capability to patients in healthcare systems that the prevailing biotech model has structurally excluded. The mission is operational rather than aspirational: affordability and global accessibility shape every technical decision the team makes.

Operating values. The team resists the prevailing biotech pattern of building diagnostics for premium-priced developed-market deployment. Cross-functional engagement is required of every team member: research scientists touch clinical partnerships, clinicians touch the data infrastructure, and engineers touch the science.

Founding principles. The consortium's work is structurally connected to the question of how a stage-one cancer detection capability could be made available to a patient anywhere, including in low- and middle-income healthcare systems. Candidates whose prior work shows orientation toward globally accessible deployment — work conducted in or with LMIC institutions, designs for resource-constrained settings, open-source release of research artefacts — are particularly well-suited to the team's culture.

Per-candidate component scores

Aisha — score 8.50 (rank 1)
ComponentP(X=1)
Education0.99
Eligibility0.93
Technical skills0.95
Domain0.97
Validated output0.85
Environmental pedigree0.60
Trajectory0.92
Cultural alignment0.95
Léa — score 5.46 (rank 2)
ComponentP(X=1)
Education0.95
Eligibility0.95
Technical skills0.62
Domain0.93
Validated output0.78
Environmental pedigree0.78
Trajectory0.75
Cultural alignment0.40
Eleni — score 5.33 (rank 3)
ComponentP(X=1)
Education0.97
Eligibility0.93
Technical skills0.92
Domain0.90
Validated output0.96
Environmental pedigree0.97
Trajectory0.60
Cultural alignment0.25
Yuki — score 5.24 (rank 4)
ComponentP(X=1)
Education0.99
Eligibility0.95
Technical skills0.72
Domain0.92
Validated output0.92
Environmental pedigree0.90
Trajectory0.65
Cultural alignment0.25
Anders — score 4.80 (rank 5)
ComponentP(X=1)
Education0.99
Eligibility0.95
Technical skills0.78
Domain0.30
Validated output0.92
Environmental pedigree0.97
Trajectory0.55
Cultural alignment0.30
The API

An AI hiring platform, configured per organisation

The Job Spec Best Candidate objective exposes one endpoint. Bearer-token authentication required. The endpoint takes a position specification, an organisation context, a configuration mapping each of the eight components to an importance level, and a list of candidates. It returns a ranking with per-component probability estimates and a one-sentence evidence cite for each.

POST /v1/objectives/job-spec-best-candidate/rank

Submit a job specification, an organisation context, a configuration, and up to ten candidate profiles. Returns the candidates ranked by composite score, with per-component probability estimates and one-sentence evidence cites per estimate.

Configurable candidate ranking · Multi-criteria evaluation · Transparent candidate scoring

Latency 60–120 seconds
one LLM call per candidate, parallelised Auth Bearer token
Technical Specification
POST  /v1/objectives/job-spec-best-candidate/rank

Request Body

{
  "brief": {
    "position_specification": "string",         // required
    "organisation_context": "string"            // optional, max 50000 chars
  },
  "configuration": {
    "education":              "non_negotiable|very_high|high|moderate|low|null",
    "eligibility":            "non_negotiable|very_high|high|moderate|low|null",
    "technical_skills":       "non_negotiable|very_high|high|moderate|low|null",
    "domain":                 "non_negotiable|very_high|high|moderate|low|null",
    "validated_output":       "very_high|high|moderate|low|null",         // not gateable
    "environmental_pedigree": "very_high|high|moderate|low|null",         // not gateable
    "trajectory":             "very_high|high|moderate|low|null",         // not gateable
    "cultural_alignment":     "non_negotiable|very_high|high|moderate|low|null"
  },
  "candidates": [
    { "id": "string", "profile": "string" }                // 1-10 candidates per call
  ]
}

Components set to null (or omitted) are excluded from the formula entirely. validated_output, environmental_pedigree, and trajectory cannot take non_negotiable — they are matters of degree, not yes/no disqualifiers, and the API rejects requests that gate on them.

Response

{
  "request_id": "uuid",
  "ranking": [
    {
      "candidate_id": "string",
      "rank": 1,
      "score": 8.503,
      "components": [
        {
          "name": "education",
          "p": 0.99,
          "evidence": "MSc Computational Biology, LUMS, distinction, valedictorian."
        }
        // ... eight components total per candidate
      ]
    }
  ],
  "diagram_id": "string",
  "diagram_version": "string",
  "model_used": "string"
}

The composite score is computed by the gated additive form: the product of the gated component probabilities, multiplied by the weighted sum over the remaining contributors. Each per-component p is a calibrated probability in [0, 1], paired with a one-sentence evidence string drawn from the candidate's profile.

Example Request

# Rank two candidates against a Build-the-platform configuration
curl -X POST https://api.utilitymax.io/v1/objectives/job-spec-best-candidate/rank \
  -H "Authorization: Bearer $UTILITYMAX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "brief": {
      "position_specification": "Doctoral researcher, plasma proteomic biomarkers ...",
      "organisation_context": "Small well-funded biotech in build mode ..."
    },
    "configuration": {
      "education": "non_negotiable",
      "eligibility": "non_negotiable",
      "technical_skills": "very_high",
      "validated_output": "very_high",
      "domain": "high",
      "environmental_pedigree": "high",
      "trajectory": "moderate",
      "cultural_alignment": "low"
    },
    "candidates": [
      { "id": "eleni",  "profile": "MSc Molecular Biotechnology ..." },
      { "id": "anders", "profile": "MSc Bioinformatics ..." }
    ]
  }'
Configuration vocabulary

Five importance levels

  • non_negotiable gate Multiplicative gate. The component's probability multiplies into the composite score directly. A low gate probability collapses the candidate's overall score regardless of strength elsewhere. Available only for education, eligibility, technical_skills, domain, and cultural_alignment.
  • very_high w = 4 Additive contributor with the highest weight. Used for the dimensions the organisation will not compromise on but does not want to gate.
  • high w = 2 Additive contributor with elevated weight. Material to the ranking but not dominant.
  • moderate w = 1 Additive contributor at unit weight. The default for dimensions the organisation considers relevant but not differentiating.
  • low w = 0.5 Additive contributor with reduced weight. Included so the dimension is not entirely silent in the score, but kept from outweighing the priorities.
  • null excluded The component is excluded from the formula entirely. Equivalent to omitting the key.

The constraint that validated_output, environmental_pedigree, and trajectory cannot be gated is structural: they describe matters of degree (how much, how strong, how clearly) rather than binary qualifiers. Gating on them would force the framework to make a yes/no judgement on continuous evidence — defeating the calibration. Submitting non_negotiable for any of these three returns 422.

Status codes & errors
StatusMeaning
200Ranking computed successfully
401Missing or invalid API key
413Request body exceeds 1 MB
422Request body failed schema validation (includes attempts to gate a non-gateable component)
429Rate limit exceeded (30/min for rank)
502Upstream LLM call failed or returned malformed output
500Unhandled server error

All error responses follow the format { "detail": "message" }. The request_id is included in successful responses and visible server-side for failed calls — including it when contacting support enables exact log lookup. The lower rate limit (30/min versus the 60/min on per-document endpoints) reflects that each call processes multiple candidates in parallel.

Frequently Asked

AI recruitment tools, configurable candidate ranking, and the UtilityMax API

What are AI recruitment tools?

AI recruitment tools are software systems that use language models or other machine-learning components to score, rank, or shortlist candidates against a job specification. Most apply a fixed ranking logic determined by the vendor.

UtilityMax is a configurable AI recruitment tool: it decomposes candidate fit into eight calibrated components and lets each organisation choose which components gate the decision and which contribute additively, so the same candidates produce different rankings under different organisational priorities.

How is configurable candidate ranking different from generic AI screening?

Generic AI screening produces a single score per candidate using a ranking logic the vendor has fixed in advance. Configurable candidate ranking exposes that logic as a parameter the customer sets per role.

In UtilityMax, the eight components and their definitions are fixed; the importance level assigned to each component — non-negotiable, very high, high, moderate, low, or excluded — is set by the customer in the configuration object on every API call. The dual-configuration demo above shows the practical consequence: same five candidates, same role, two configurations, two different rankings.

Does UtilityMax offer transparent candidate scoring?

Yes. Every ranking returned by the API includes per-component probability estimates and a one-sentence evidence cite drawn from the candidate's profile for each estimate. The composite score is computed by an explicit gated additive utility function whose form is published in the methodology paper.

Reviewers can trace any candidate's composite score back to its eight component probabilities and the evidence supporting each. Composite scores are comparable within a single response (same configuration, same diagram version) and not directly comparable across configurations.

Is there an AI recruitment software API I can integrate with?

Yes. The Job Spec Best Candidate objective exposes a single endpoint, POST /v1/objectives/job-spec-best-candidate/rank, with bearer-token authentication. Each call accepts a position specification, an organisation context, a configuration mapping each of the eight components to an importance level, and up to ten candidate profiles.

Typical latency is 60 to 120 seconds. The API is currently offered in trial with selected partners — see the request form below.

Does UtilityMax replace human hiring decisions?

No. UtilityMax produces a ranking with explicit component probabilities and evidence cites that hiring teams use as input to their own evaluation. The component probabilities and evidence cites exist precisely so that a human reviewer can interrogate any score before acting on it.

Final hiring decisions remain a human responsibility, supported by interviews, reference checks, and the organisation's existing review processes.

How does the framework handle bias in candidate ranking?

The framework's design choices are aimed at reviewability rather than at any guarantee about bias. Each of the eight components has an explicit, observable definition; each per-candidate probability is paired with a one-sentence evidence cite from the candidate's profile; and the configuration that determined the weighting is part of the audit trail.

This means a hiring team — or an external reviewer — can inspect why a given candidate ranked where they did, which component drove the score, and what evidence supported each estimate. Bias considerations remain the customer's responsibility within their applicable employment law framework; the framework's transparency is intended to support that work, not to replace it.

Request Trial Access

Run UtilityMax against your own role

Trial access is currently available to selected partners. Tell us about the role you want to evaluate — the position specification you would normally post, the kind of candidates you typically see, and the dimensions that matter most for your organisation. We will respond within one business day with next steps and a link to schedule a 30-minute onboarding call.

Trial request · ~1 minute

Apply for trial API access

We use the information below to construct an organisation context and configuration together on the onboarding call. Your details are processed in line with POPIA.

We respond within 1 business day · No marketing emails

By submitting, you consent to UtilityMax processing your details for the purpose of evaluating and supporting trial access, in line with the Protection of Personal Information Act (POPIA). Trial outcomes vary by role, candidate pool, and configuration; UtilityMax does not guarantee any specific hiring outcome on any individual role.

Or Reach Out Directly

If you'd rather skip the form, write directly. Trial integrations begin with a 30-minute onboarding call where the organisation context and configuration are constructed together and the first ranking run is walked through end to end.

Ofir Marom PhD Computer Science · Published at NeurIPS, AAAI, ICAPS · 14 years industry experience
ofir@utilitymax.io Johannesburg, South Africa  ·  utilitymax.io