CV analysis: your criteria applied to every application
A vacancy sets out precise requirements: qualification, years of experience, skills, languages, mobility. Your agent reads each application, finds those elements and states, for each one, what the application explicitly mentions. Where you have set objective criteria — including exclusion criteria, such as a minimum number of years of experience — it applies them and presents the applications sorted accordingly. Hosted in France: your candidates' data stays with you. The recruiter decides who is interviewed.
Updated on
For each application, what the file explicitly mentions on each of these points.
Your exclusion criteria are applied and the resulting split is presented to you.
🔗 Sourced · elements taken from the applications
Where the application is silent on a requirement, it is marked as such — never inferred.
Choosing who is interviewed is yours: it concerns people.
✎ Support · elements presented, the recruiter's decision
A Blue Lemon Agent CV analysis agent takes the requirements from your vacancy, states for each application what the file explicitly mentions and applies the objective criteria you have set, exclusion criteria included. A point not mentioned is marked as such, never inferred, and no ranking of people is produced. It runs on local inference or is hosted in France: your candidates' data stays with you, architecture designed to reduce exposure to extraterritorial legislation, location alone not being enough to guarantee immunity.
These figures describe our offer, not results measured at a client. How large the gain is on your volume of applications received is confirmed by a pilot.
What does an AI agent bring to the analysis of your applications?
Finding in every application the elements a vacancy requires is methodical reading. That is exactly what an agent does well, provided it sticks to the facts.
! The issue
A vacancy expresses requirements that can be verified: qualification, length of experience, languages, mobility. Finding them in a hundred and twenty applications means long and steady reading. The agent takes it on and returns for each application what the file explicitly mentions, requirement by requirement, with the corresponding wording to back it up.
✓ Our answer
Where you have set objective and lawful criteria — a minimum number of years of experience, a mandatory authorisation, a required certification — the agent applies them and presents the applications sorted by those rules, each one justified by the wording in the file. Choosing who is interviewed remains yours: the European AI Act classes recruitment among the high-risk uses, and a decision about a person calls for a human eye. Local inference or an isolated resource hosted in France: your candidates' data is entrusted to no third party.
Candidates' personal data: sovereignty & compliance
Applications are personal data, often sent for a single recruitment. Here is how the architecture of our agents protects them.
Local inference
The agent can run on a machine belonging to your organisation: no application and no candidate data leaves the network.
Hosting in France
Otherwise, a dedicated and isolated resource hosted in France, under French law — your vacancies and your applications: processing and access within the European Union targeted by the architecture.
Reduced extraterritorial exposure
For candidates' personal data, the architecture aims to reduce exposure to the Cloud Act and FISA 702; being located in France or in the European Union does not, on its own, guarantee immunity.
Isolated resource
No pooling: an environment strictly dedicated to your company and its vacancies.
Every assessment refers back to a mention in the file
A requirement not mentioned is marked as such rather than inferred; encryption, role-based access, logging and purging at the retention deadline you set.
AI Act: governed deployment
The agent is strictly in support; no candidate is set aside or taken forward without the recruiter's decision; traceability and human oversight from end to end.
What depends on the architecture chosen These points are not general guarantees: they are settled deployment by deployment, in the quotation.
- The applicable location is that of the architecture set out in the quotation and verified before commissioning.
- Local execution is announced only for the configuration explicitly described and accepted in the quotation.
- The applicable isolation depends on the deployment mode set out in the quotation; no dedicated isolation is presumed.
- Roles and permissions are configured and accepted for the identities and systems actually connected.
- The events logged, their content, their retention period and who may access them are defined for the deployment chosen.
See the agent at work
4 real situations, taken from those that come up most often. Pick one: the exchange unfolds as it would in your organisation.
A scripted demonstration. These exchanges show how the agent behaves — its sources, its refusals, what it leaves to your teams. Nothing is sent from this page, no model is queried here, and the matters named are fictional. That is precisely what we promise your data.
The behaviours shown here — monitoring, automation rules, routing and reminders — are configured with you during deployment, from your tools, your rules and your thresholds.
The architecture points named in these exchanges — location, local execution, isolation, encryption, role-based access, logging — are not a guarantee attached to the demonstration: they are those of the architecture set out in your quotation, and verified before commissioning.
· Your logistics manager advert states 14 requirements. Six are verifiable on no CV — autonomy, team spirit, rigour, adaptability, initiative, interpersonal ease.
· It asks for 12 years of experience. The three people who have held the post here had 4, 6 and 7 on appointment.
· One requirement is marked "essential" and is not, in practice: across your last six recruitments of this kind, four people did not meet it.
· Eleven employees hold a qualification or certification that appears nowhere in your internal vacancies. They never see one go by. morning-watch_4-flags.pdf3 of 4 flags concern the adverts
⛓ Source · published adverts, recruitment history, personnel files, internal vacancies
What the advert asks: "minimum 12 years' experience in logistics".
What your files show: the post has been held by three people since 2014. On appointment: 4 years, 6 years, 7 years. None would have passed your own advert's filter.
What I do not conclude: that the requirement is wrong. The post may have changed, and so may expectations, and a requirement can be deliberately raised.
What I do record, and it is measurable: of the 214 applications received, 168 state fewer than 12 years. If the figure is applied, it sets aside 78% of the pool; if it is not, it discourages people who never apply.
What I propose: that the figure be decided, whatever it is. Today it is in the advert and applied nowhere — the one situation with no advantage at all. 12-years-asked_4-6-7-observed.pdf168 of 214 CVs below the threshold
⛓ Source · published advert, 3 postholder files since 2014, 214 applications
Routing follows whoever writes the advert: an unverifiable requirement goes to the drafter, before publication — afterwards it has already filtered; a gap between advert and actual recruitments to the recruitment lead; an "essential" requirement not applied to the same person, with the six files; an employee whose skills cross no vacancy to internal mobility, never to their manager.
With a chase: before publication on an advert, 7 days otherwise. Then a monthly summary: by advert, never by candidate.
What this morning has already earned you: 4 flags, 3 of them before an advert filtered anyone at all. From tomorrow that means 6 unverifiable requirements removed before publication, a 12-year threshold decided rather than endured — 168 CVs out of 214 hang on it — and 11 employees who finally see the posts made for them go by. The access stays yours: opened by role, logged, withdrawn with a word, and nothing leaves your walls. The decision to interview or set aside belongs to the named recruiter, and I hand it back in minutes: every application arrives with the file complete, every assessment points to the line of the CV behind it, and all 214 go through the same grid — equal treatment is written down, dated, and can be demonstrated if a decision is challenged. The next step is ready: give me the logistics manager advert before it is republished, and I will hand back the corrected version and the 214 applications re-sorted on the criteria you settle.
✎ Framework · a deliberate asymmetry — offers rank, people do not
Ranking offers for a person: the person receives a list, reads it, ignores it if they wish, applies to the twelfth if they want. A bad ranking costs them five minutes.
Ranking people for an offer: the person does not see the list, does not know they are on it, and if they are near the bottom nobody will open their file. A bad ranking costs them the post, and they will never know.
What differs between the two: who decides, and who can contradict the result. In the first case, the person concerned. In the second, nobody.
So what I do: for an employee seeking mobility, the internal vacancies that cross what they have declared, with the shared word highlighted. For a vacancy, nothing resembling an order.
This is not token caution: the same engine serves both directions, and it is wired in only one. two-directions_one-wired.pdfWho decides · who can contradict
✎ Framework · the same engine, wired in one direction only
What I record: eleven employees declare a qualification or certification — in their file, with their agreement — that appears in none of your 34 internal vacancies of the last twelve months.
What that may mean, and I do not settle it: either you have no matching post, or your internal adverts use a different word from the certification's. Of the eleven, seven concern a certification with a commercial name; your adverts describe the skill in plain words.
What I propose: for those seven, show both wordings side by side — the certification's and the advert's — without deciding they mean the same thing. That is for the employee to say.
Who the list goes to, and why that is the gain: each of the eleven receives their own, with the vacancies that describe their skill in plain words and the wording of their certification alongside. Of your 34 vacancies over the last twelve months, 9 describe a skill held by one of the eleven — that many internal applications nobody ever had the chance to receive. Applying stays their own move, one click in the message, and that is exactly what makes the application credible: the recruiter who receives it knows it was wanted. 11-employees_0-crossing-vacancy.pdf7 of 11: a different word, not a missing post
⛓ Source · 11 files, 34 internal vacancies over 12 months
The eight verifiable ones: a qualification, a length of experience, a sector, a named tool, a language, a licence, a certification, a geographic area. Each is found or not found, and I say which.
The other six: autonomy, team spirit, rigour, adaptability, initiative, interpersonal ease. No CV carries them. A candidate can write them — and 189 of the 214 do — which makes them all the less discriminating.
What tools do about this: they look for the words in the CV and count. That measures who wrote "autonomous", not who is.
What I do: I display them as "to be assessed at interview", and look for nothing. Six empty lines on my report, and that is honest.
What I propose to the drafter: keep them if they describe the post, but know that they sort nothing — and that the sorting will therefore happen on the other eight, decided or not. 14-requirements_8-verifiable.pdf189 of 214 CVs write "autonomous"
⛓ Source · published advert, 214 CVs received, occurrences recorded
What the advert says: a named certification, "essential".
What your last six recruitments on equivalent posts show: two people held it, four did not — two of whom obtained it after joining.
What that produces today: of the 214 applications, 163 do not mention that certification. A candidate reading it as disqualifying does not apply — and those people I do not see and cannot count.
What I do not say: that it should be removed. It may have become genuinely essential, and your past recruitments do not bind you.
What I propose: two words instead of one — "essential" or "desirable" — and that the choice be made. An "essential" applied one time in three is not a requirement: it is a sort whose rule nobody knows, neither the candidates nor the people recruiting. 1-essential_4-hired-without.pdf163 of 214 CVs without the certification
⛓ Source · published advert, 6 equivalent recruitments, 214 applications
On the left: the advert's eight verifiable requirements, as it writes them.
On the right: what the CV says, with the page and line, or "not found".
What is not on it: a percentage, a count of requirements met, a colour, an order, a conclusion. None of those objects is produced.
Why not even the plain count "6 requirements of 8": because it adds up things that do not add up. A missing licence and a missing qualification do not weigh the same, and the count pretends they do.
The report's most useful line, and its most discreet: where the CV and the advert use two different words for what might be the same thing, I show both, side by side, without saying they are equivalent.
Across the 214 CVs: 67 use a different word from the advert on at least one requirement. A tool looking for the exact word sets all 67 aside; a tool deciding they are equivalent wrongly matches some. Showing both leaves the question to whoever reads. 2-columns_0-percentage.pdf67 of 214 CVs: a different word
⛓ Source · published advert, 214 CVs, wordings recorded
What is kept: each advert's requirements and whether they are verifiable, gaps between advert and actual recruitments, pairs of differing wordings — the advert's word and the CV's, without names — and internal vacancies with no matching employee.
What is not kept: no CV report beyond the close of the recruitment, no score or ranking — they do not exist —, no candidate profile, no skill extracted from a CV and poured into a database, and no employee's viewing of internal vacancies.
The last matters more than it looks: knowing who looks at internal vacancies is knowing who is considering leaving their department. That information is neither kept nor reported, not even as a count.
What the monthly summary contains: unverifiable requirements by advert, "essential" requirements not applied, the most frequent wording gaps, and internal skills crossing no vacancy. Four indicators about your adverts. what-is-kept.pdf4 items kept · 5 impossible
✎ Framework · retention periods to be set by the company
Every line of the report carries four things: the requirement as the advert words it, the sentence from the application quoted word for word, the page and line where it sits, and the kind of document — CV, covering letter, attachment. From the logistics manager advert: "Requirement: master's degree in logistics — mention: 'MSc Logistics and international procurement, 2016' — CV, page 1, line 6."
Where the statement does not exist, the cell says "not mentioned in the application", and never anything else. Across the 214 applications received for your six open adverts, 1,842 findings were produced and 391 carry that wording — 21%. It is a high figure, it is accurate, and above all it tells you that your adverts ask for things candidates do not think to write down.
Why the reference changes everything, including for you: a candidate who challenges a finding gets the sentence from their own application and the line it was read on. A finding that can be traced to a statement can be argued with; a finding with no reference cannot be argued with, only endured — and it is you it exposes.
What traceability turned up against me: 23 references pointed at the wrong page, all on CVs laid out in two columns, where the line read was not the line displayed. No finding was wrong, but a wrong reference is as good as no reference: the reader finds nothing and stops trusting the others. Two-column documents are read differently since; across the following 96 applications, your recruitment officer checked the references one by one, with no discrepancy.
And the rule of the report does not move: two columns, no scores, no ranking of people. Traceability exists to ground what is written, not to grade anyone.
✎ Framework · traceable findings — one mention, one page, one line
Your case is not here? That is exactly what a 15-minute conversation is for. Book the free audit →
What does the agent actually do?
One agent, several moments in the recruitment process. All these uses work in support, subject to your approval.
Reading the applications
Finds in each file the elements the vacancy requires.
Applying your criteria
Applies the objective criteria you have set, exclusions included.
Traceability of assessments
Refers every assessment back to the wording in the file it rests on.
Need to go further?
These agents handle a different business process, with their own owner and their own price. They are added to this one.
In 15 minutes we identify the most relevant agent — without oversizing the project.
How much time can a recruiter give back to interviews?
By taking on the reading requirement by requirement, the effort shifts towards meeting the candidates. How large the gain is depends on your volume and remains to be confirmed by a pilot.
The stages of your AI agent project
Audit & scoping
15 minutes to target the use case with the best return.
Quote or direct sign-up
A catalogue offer is bought online; a specific need gets a costed quote.
Design
We design the agent and its guardrails.
Integration & testing
We connect your tools to the agent, which is itself hosted in France.
Rollout
Going live and training your team.
Operation
Continuous supervision and improvement.
One package, one agent
A CV analysis agent (reading, criteria, traceability), installed and operated for you.
Setup + controlled subscription
- Installation, configuration and training for your teams
- Operation, human oversight, updates and support
- Sovereign hosting in France, a dedicated and isolated resource
All inclusive, no setup fee
- Setup included (installation, configuration, training)
- Operation, human oversight, updates and support
- Sovereign hosting in France, managed end to end
On site, you own it
- Hardware installed on your premises (you own it)
- French / European AI models run locally
- Secure remote maintenance (Pro support included)
Four guarantees that matter in recruitment
Related resources
Your questions, our answers
Does the agent rank candidates?
Can it be given exclusion criteria?
What does it do if a file does not mention a requirement?
How is the European AI Act respected?
How long are applications kept?
How long does it take to deploy this agent?
Other agents for your recruitment
Let's size up the potential in your recruitment
15 minutes to frame your vacancies and your criteria — hosted in France, supervised, with no commitment.