HelvioAI HR agent · working name
Commercial proposal
Prepared by Raccoon Gang
21 June 2026 · v1 · draft for discussion
Proposal

An AI recruiting agent for Swiss companies, built where the data has to stay.

You asked us to scope an Android app with an AI agent for HR: CV parsing, candidate-to-vacancy matching, interview-question generation, sign-in through Google Identity, Swiss data residency, and links to LinkedIn, the major CRMs, and the HR systems Swiss SMEs run. Here is what we would build, what it costs, and where we would tell you to spend nothing.

The short version

from USD 27k
Fixed price for a working, Swiss-resident MVP on Android (≈ CHF 24k). Expected range 27–39k by final scope.
6–8 weeks
From signed scope to an app running on your own CVs. Discovery starts at USD 2,700.
~$0.03 / CV
Running cost to parse, score and explain a CV. Swiss hosting ~$300/mo. No per-seat tax.

Numbers below use our standard blended rate of USD 60/hour and a three-point estimate. The "AI-accelerated" column is what we actually target by building with AI tooling; the "expected" column is the safety net. We would rather quote the lean number and beat it than pad it.

01

What you asked for, and our honest read

The brief lists the feature set of an enterprise talent-acquisition suite. Most of it is very buildable. Two items are harder than they look, and saying so up front is the point of this document.

Our recommendation: ship the AI core first as a Swiss-resident Android app with Personio and one CRM connected. That earns trust on real CVs in weeks. Then add the heavier integrations and iOS once the value is proven. The full brief is a phase-two-and-three conversation, not a first invoice.

02

What we would build

Phase 0 — Discovery & architecture

from USD 2,700
~1.5 weeks · often credited toward Phase 1
  • Workshops with your HR and IT leads; success metrics; a frozen, signed scope.
  • The one decision that shapes everything: may candidate data sit in an EU region (Switzerland is on the EU adequacy list since Jan 2024), or must inference run physically on Swiss soil? Each leads to a different AI stack. We decide it here, on paper.
  • Data model, security model, and a data-protection impact assessment outline (revFADP + GDPR).

Phase 1 — MVP: the AI core, on Android, in Switzerland

USD 27k–39k
~6–8 weeks · fixed price · the clickable prototype is already built (see section 07)
  • Android app (Kotlin / Compose Multiplatform) covering roles, candidates, ranked shortlist with explained scores, parsed-CV view, interview kit, and a light agent chat.
  • The AI pipeline: parse a CV to structured data, score it against the role with reasons, generate tailored interview questions — in DE / FR / IT / EN.
  • Sign-in via Google Identity Platform (multi-tenant, enterprise SSO, MFA).
  • Two-way sync with Personio and HubSpot, on a connector core that the rest plug into later.
  • Swiss hosting, encrypted storage, audit logging, consent and AI-use disclosure, data export and erasure, and human-in-the-loop on every ranking decision.

Phase 2 — Integrations & reach

modular · ~USD 25k–35k
priced per item, once Phase 1 is in real use
  • iOS from the same codebase (Compose Multiplatform): ~USD 6–8k.
  • Salesforce + Microsoft Dynamics 365 connectors: ~USD 8k.
  • LinkedIn, the compliant way — sign-in, candidate-consent import (EU/CH data portability), and a vetted public-data provider: ~USD 4k.
  • Workday / Abacus umantis: per-customer setup, time-and-materials (~USD 6–12k each), only when a deal calls for it.
  • Reporting, deeper agent, gamification: priced on request.
03

Phase 1 estimate, broken out

All hours are all-in (they include review, testing and bug-fixing). Blended rate USD 60/hour.

WorkstreamExpected (h)AI-accelerated (h)
Backend foundation, Google Identity (multi-tenant), RBAC7250
AI service — parse · match & explain · interview Q · agent chat · residency routing12084
Android app (Compose Multiplatform) — all core screens168118
Integrations — sync core + Personio + HubSpot7653
Compliance & trust — consent, audit, human-in-loop, export/erasure5639
QA, client UAT, Swiss production release6848
Project management & communication8862
Total — Phase 1648 h454 h
At USD 60/h≈ $38,900≈ $27,200

≈ CHF 34,200 / CHF 24,000 at the prevailing rate. Add Phase 0 (USD 2,700–3,800). For comparison, a recent comparable internal build of ours that also spanned AI, a mobile app, a web dashboard and multiple integrations came in around USD 132k at full scope — this proposal is deliberately scoped down to the part that proves value first.

04

What it costs you to run

05

The honest risk read

Where the brief meets reality. We would rather you hear this from us now than from a lawyer later.

Green

CV parsing, matching, interview questions, Swiss hosting, Google sign-in

Solved territory. LLM-first parsing and scoring is multilingual and cheap; Swiss-resident hosting has several real options; Google Identity Platform does multi-tenant SSO out of the box.

Yellow

Google Identity stores identity data in the US

The sign-in service itself processes a small set of identity fields in the US, with no region setting. Minor PII, but we flag it for your DPO and, if strict on-soil residency is required, swap in an EU-region identity provider. A discovery decision, not a surprise.

Yellow

Salesforce / Dynamics / Workday integrations

All doable. Salesforce needs an event-stream subscriber rather than a simple webhook; Dynamics needs tenant-side setup; Workday is a per-customer provisioning project. We build a connector core so each is an adapter, not a rewrite.

Red

Bulk LinkedIn profile / article import

There is no sanctioned way to pull arbitrary LinkedIn profiles — LinkedIn's API only returns the signed-in person's own data, and it actively sues scrapers (it shut Proxycurl down in 2025). We build the compliant path instead: sign-in for identity, candidate-consent import via EU/CH data portability, CV upload, and a vetted public-data provider with contractual cover. Honest, and it keeps the client out of trouble.

Yellow

EU AI Act — recruitment AI is "high-risk"

Filtering and ranking candidates is explicitly high-risk, and revFADP puts the duty on a named person. We build for it: a human signs off every decision, scores come with explanations, and everything is logged. This is a selling point for a Swiss buyer, not a blocker.

06

The stack we would use

Mobile
Kotlin Multiplatform + Compose MultiplatformAndroid now, iOS later from the same code. Stable for iOS since May 2025.
Backend
Python · FastAPI · LangGraphThe AI ecosystem lives in Python. A Go service can take any hot path later if load demands it.
AI
Claude, via an EU regionParse, match and draft in one pipeline. Run through AWS Bedrock / Vertex EU, or a Swiss-hosted model if data must stay on-soil.
Data & hosting
Postgres + pgvector, in SwitzerlandExoscale or Infomaniak — Swiss-owned, Swiss-law. Encrypted at rest, CV files in Swiss object storage.
Auth
Google Cloud Identity PlatformMulti-tenant, SAML/OIDC SSO, MFA, token verification on the backend.
Integrations
Connector core + per-system adaptersPersonio & HubSpot first; Salesforce, Dynamics, umantis, Workday as adapters.
07

See it before you sign anything

The prototype below is real and clickable — every screen in the Phase 1 app, built to the standard it would ship at. The three landing pages are one product, three design directions, so you can pick a voice.

08

Beyond this one project

We understand there may be three to five AI builds in the pipeline. The same approach travels: a tight discovery, a lean first build that proves value, then phased expansion. For planning, our blended rate is USD 60/hour, with the AI-accelerated delivery passed back to you as fewer hours rather than a higher rate. We are happy to scope each new brief the same way we scoped this one — including telling you which parts not to pay for.