/en/agents · automation
Iconicompany agent analyzes vacancies and resumes, matches specialists, verifies requirements, and tracks candidate responses. Workflow, iagent integration, checks, and boundaries of autonomy.
italent is an Iconicompany product for talent acquisition, part of its agent platform. It receives job openings and resumes, finds candidates, matches their experience against requirements, drafts invitations, and manages candidate responses up to client handover.
The result is used by recruiting managers, client companies, staffing agencies, and specialists themselves. Managers receive candidates with a clear breakdown of qualifications. Specialists see the job opening, clarify their experience and working terms, accept the invitation to talk, or decline.
The product includes a portal, recruitment services, and agent tools. Chat communication in messengers is handled by the dedicated iagent product: it delivers messages and triggers italent actions. Access rules, match scoring, and candidate status tracking remain inside italent.
Recruiting starts with a vacancy and available specialist data:
A vacancy can come from the portal or via the iagent gateway. Along with the message, the gateway forwards the source and discovered links, including button targets and hidden hyperlinks. Missing links to external job portals are never guessed.
Completeness of a resume affects the result: the absence of a stated skill in the text does not necessarily prove lack of ability. Missing information must be clarified directly with the specialist.
The system stores vacancy and candidate information, extracting requirements, skills, and experience. The language model helps parse unstructured text; downstream search utilizes structured data and semantic embeddings.
The link to the vacancy source is preserved. For chat-driven actions, the system identifies the operating user and validates their permissions.
Candidate matching queries available talent records and applies configured screening rules. Preliminary semantic similarity is tracked separately from strict requirement fulfillment.
For each "vacancy — specialist" pair, a match record is created (referred to internally as an offer). Here, "offer" means a candidate pipeline card; its creation does not imply an employment offer or candidate agreement.
Re-running matching does not generate duplicate offers for the same pair. Early filter dropouts are recorded with reasons and do not clutter the pipeline by default.
Verification builds a checklist: which requirements are confirmed by candidate data, what is missing, and what requires clarification. Skill ontology rules and the model assist in parsing.
The aggregate score is calculated deterministically from the checklist with mandatory and optional weightings. It is not requested as an arbitrary number from the model. Adjusting threshold criteria does not require re-evaluating candidate experience from scratch.
Ontology modifications are handled under a distinct review procedure. Processing a vacancy can strip noisy keywords locally, but cannot automatically declare one skill as evidence of another across the platform. Such decisions require an authorized auditing agent or human review.
Based on match scores and settings, the system rejects the offer, keeps it for later review, or invites the candidate to clarify details. Automated invitations can be toggled independently of sourcing; recruiters also retain manual control.
Dispatch respects available channels, opt-out preferences, and quiet hours. Delivery status is logged: missing contact info, channel opt-outs, and send failures have distinct codes.
In dialogue, the agent distinguishes between interest, refusal, and clarification questions. Recording an answer via iagent requires quoting the candidate's last message; model paraphrasing is not accepted. Salary questions are logged in history without marking acceptance. Expressing interest requires expected compensation and collaboration format; missing parameters are requested directly from the person.
In adaptation workflows, candidates connect stated requirements with specific entries in their background. The system prepares a tailored copy of the resume that can be reviewed and edited prior to submission.
Factual verification remains the candidate's responsibility. Model-generated copy requires verification and does not serve as standalone evidence of hands-on tool experience or project ownership.
Responses with agreed terms proceed to submission. If resume tailoring was triggered, dispatch waits for its completion; unhandled errors must never be masked by sending the raw document.
Submission mechanisms vary by client: internal work chats, bot dialogues, or external ATS forms. Each route requires dedicated credentials and validation. The presence of an adapter does not imply universal or automated access to every third-party system.
| Artifact | Contents |
|---|---|
| Vacancy Card | Requirements, terms, hiring organization, and source metadata |
| Specialist Profile | Resume, employment history, skills, and available contacts |
| Offer (Pipeline Card) | Link between candidate and vacancy, checklist, status, and response terms |
| Communication History | Invitations sent, candidate responses, and delivery logs |
| Adapted Resume | Tailored resume version open for inspection and manual edits |
| Handover Status | Recorded submission event or exact reason for wait/rejection |
Managers can clearly differentiate unreviewed candidates from vetted or rejected ones. Candidate replies, resume readiness, and delivery status are tracked separately.
iagent receives incoming messenger traffic, conveys vacancies and resumes, displays pipeline cards, and triggers actions on behalf of authenticated users. Tools are invoked via MCP or preconfigured action buttons without unnecessary model steps.
italent verifies permissions and executes domain operations. Gateway credentials do not grant broad access to unauthorized vacancies or candidate data.
An auditing agent reviews disputed skill mappings and ontology updates, separating global rules from single-candidate evaluations.
idevops manages service infrastructure, dependency health, and operational diagnostics, without involvement in recruiting decisions.
Models may misinterpret domain requirements, confuse adjacent technologies, or extrapolate unsupported experience. Incorrect global skill associations are especially risky because they skew scoring across multiple candidates.
Outdated resumes, incomplete contact details, and unverified job-seeking intent limit search effectiveness. High match scores do not guarantee candidate engagement or successful hiring.
Third-party platforms may alter forms, restrict access, or reject submissions. Automatic submission cannot be guaranteed universally.
Personal data handling must comply with consent parameters. Tooling and storage deployments follow organizational boundaries without unconditional claims of purely local processing.
Workflow example: A recruiter posts a vacancy, the system matches a specialist and sends an invitation. The candidate asks: "What is the compensation range for this position?"
iagent passes the clarification intent and the quoted message to italent. The product verifies the user, offer access, and quote integrity, logging the question in history. The offer status remains unchanged: an inquiry is not an agreement.
Later, the candidate confirms interest, stating expected compensation and engagement format. The system records the response. If resume adaptation was requested, submission waits for completion.
Verification: History records the exact question without altering state; the subsequent response records candidate terms. Document readiness and submission success are validated independently.
| Metric | Measurement Method |
|---|---|
| Time to Relevant Response | From vacancy intake to candidate reply approved by manager |
| Selection Quality | Ratio of qualified candidates among suggested and missed profiles |
| Checklist Accuracy | Erroneous vs missed requirement validations against ground truth |
| Response Rate | Share of delivered invitations with responses (questions vs interest vs declines) |
| Handover Success | Ratio of agreed submissions delivered to recipient and failure reasons |
| Manual Corrections | Number of human edits applied to requirements, scores, resumes, or terms |
Valid comparisons require comparable vacancies, standardized quality benchmarks, and accounting for all manual recruiter effort. Match scores do not equate to hiring probability. Time savings and increased placement rates require dedicated empirical verification.