Scope — Annex 3 Deliverable → Feature Mapping
Source: grant/scope.md · synced 2026-06-12
Every feature we build must trace back to one of the 3 Annex 3 deliverables. This is the authoritative scope doc. If a feature isn’t here, it’s out of scope.
Source: Letter of Offer Annex 3 (Project Milestones and Deliverables), Case Ref 26032AYH. Approved framing (from Annex 4): “AI Platform that helps AI Candidate Sourcing and Matching Automation. This will be a product that we will be marketing to other Headhunting Agencies.”
The 3 Annex 3 Deliverables
1. AI-Powered Resourcing
Definition (ours): Automated identification and aggregation of (a) candidate profiles and (b) hiring employers, using public data sources, with structured filtering and AI-driven enrichment.
In-scope features:
| Feature | Internal value | Audit-facing value |
|---|---|---|
| Candidate scraping (LinkedIn, public profiles) via Apify | Calibre’s primary sourcing engine | Demonstrates “automated candidate sourcing across multiple platforms” |
| Job listing scraping for BD (LinkedIn Jobs, careers pages) | Identifies new BD targets | Demonstrates “AI-Powered Resourcing” extends to employer-side |
| Apollo integration for company + decision-maker enrichment | Deep BD intelligence | Productisable as multi-tenant data enrichment |
| Anymail Finder for verified email addresses | Outreach deliverability | Same |
| AI deep research per candidate/company (LLM with web context) | Surfaces relevant insights | Demonstrates “intelligent identification” claim |
| Structured data filters (industry, years, role) | Filter longlists fast | Configurable per tenant for productisation |
| Tagging engine (auto-tag candidates: skills, certs, languages) | Builds tagged talent pool | Per-tenant taxonomies |
Out-of-scope: Live job-board API integrations (LinkedIn Recruiter, Indeed paid APIs), candidate self-signup portal.
2. AI Candidate Filtering & Matching
Definition (ours): Domain-specific evaluation and matching system that ranks candidates against JDs using human-validated conditional rubrics + composite scoring.
In-scope features:
| Feature | Internal value | Audit-facing value |
|---|---|---|
| JD parser (extract role, skills, requirements from uploaded JD) | Speeds up intake | Demonstrates “AI” in the matching pipeline |
| Conditional Mapping Logic engine (recruiter-validated rules) | Calibre-specific rubrics for each vertical | Configurable rubric system per tenant — productisable |
| AI Scoring Engine (weighted, vertical-specific criteria) | Quantitative ranking | Demonstrates “scoring” claim |
| Composite ranking + shortlist generation | Recruiter-facing top-N output | Same |
| Encyclopedia of vertical terms (Commodities, Maritime, Offshore, Tech & Gaming) | Domain knowledge baked in | Pre-loaded vertical packs sold to other agencies |
| Title normalisation + skills adjacency mapping | Catch non-obvious fits | Adds depth to the “intelligent” matching claim |
| Two-layer filter: hard fit (score <4 drop) → deep assessment | Structured shortlist | Demonstrates rigor |
Out-of-scope: Real-time interview scheduling, video screening, behavioural assessments, ATS-style pipeline management.
3. AI-Driven Candidate Outreach
Definition (ours): Personalised, multi-channel outreach generation and orchestration, with automatic database enrichment from responses.
In-scope features:
| Feature | Internal value | Audit-facing value |
|---|---|---|
| AI Personalisation Engine (deep research → tailored email subject + body) | Higher response rates | Demonstrates “deep personalisation” claim |
| Tone guide / prompt-engineered templates per vertical | Calibre’s brand voice | Configurable per tenant |
| Email send (via existing infra — Instantly/Smartlead) | Delivery + tracking | Same |
| LinkedIn touchpoint sequencing (connect requests + InMail copy) | Multi-channel coverage | Productisable |
| Auto-add responders to tenant’s database | Builds tagged talent pool | Per-tenant DB |
| Auto-tagging by response signal (open/responsive/disinterested) | Data hygiene | Same |
| BD outreach to hiring employers (thin module) | Wins new mandates | Maps to “AI-Powered Resourcing” employer-side |
Out-of-scope: SMS, WhatsApp Business API, voice/calling automation, real-time chat.
Cross-Cutting Features (for productisation defensibility)
These don’t belong to a single Annex 3 module but are required to make the system defensibly a “product marketable to other Headhunting Agencies.”
| Feature | Why it’s required |
|---|---|
| Multi-tenant DB (Supabase RLS) | Fundamental for “marketable to other agencies” |
| Public marketing landing page | Demonstrates productisation intent |
| Signup → email verification → onboarding wizard | Demonstrates customer acquisition flow |
| Login + auth (Supabase Auth) | Demonstrates real product login |
| Stripe checkout (test mode) + pricing tiers | Demonstrates monetisation |
| Admin dashboard for tenant management | Demonstrates operational readiness |
| Configurable per-tenant settings (verticals, rubrics, tone, branding) | Demonstrates the product can serve multiple agencies, not just Calibre |
What’s Explicitly NOT in Scope
- Mobile app
- Native iOS/Android
- Real-time collaboration features
- Video interview tooling
- Background check integrations
- Payroll / contract management
- Proper analytics warehouse (basic event logging only)
- White-label theming for tenants beyond logo + colour
12-Step → Annex 3 Deliverable Mapping
The build is tracked through the 12-step system in SOURCE-OF-TRUTH.md §2. Each step traces back to one of the three Annex 3 deliverables (or to the Annex 4 productisation framing, or to the explicit Annex 3 audit deliverables themselves):
| Step | Maps to | Why |
|---|---|---|
| Step 1 — JD ingested | AD2 (Filtering & Matching — intake side) | JD intake is the first stage of “AI Candidate Filtering & Matching”; the JD-side processing must exist before any matching can run. |
| Step 2 — JD parsed | AD2 | Same — structured JD fields feed the matching engine. |
| Step 3 — JD normalised + tagged | AD2 | The controlled vocabulary is what the matching engine pivots on. |
| Step 4 — Sourcing | AD1 (AI-Powered Resourcing) | Candidate scraping + employer-side BD = the entire “resourcing” deliverable. |
| Step 5 — CV parsed + matched | AD2 | The matching engine intake side — CV parsing (Step 5.1) + CV normalisation (Step 5.2) + recruiter-validated rules (Step 5.3 industry-expert filter). Step 5.3 fires between Step 6 and Step 7 at runtime; numeric ID groups it under Step 5. |
| Step 6 — Hard-fit filter | AD2 | Layer 1 of the scoring engine. |
| Step 7 — Deep assessment | AD2 | Layer 2 of the scoring engine. |
| Step 8 — Shortlist | AD2 | Final ranking + verdict synthesis = the deliverable output of “Filtering & Matching”. |
| Step 9 — Outreach | AD3 (AI-Driven Candidate Outreach) | Personalised messaging + multi-channel sequencing = the entire “outreach” deliverable. |
| Step 10 — Productisation surface | Annex 4 framing | Multi-tenant + signup + Stripe + demo tenants = the artefacts that defend the “marketable to other Headhunting Agencies” claim. Not a billable Annex 3 deliverable; supports the framing the project was approved under. |
| Step 11 — Audit prep | Annex 3 explicit deliverables | Project report on Calibre letterhead, training materials, attendance log, productivity measurement — these are itemised separately in the approval letter as required audit deliverables. |
| Step 12 — Infra | (no AD) | Internal infrastructure (docs site, sync pipeline, dev environment) that the build needs but is not a grant deliverable in itself. |
Step 5.3 — the industry-expert filter — is a particularly strong audit signal. It directly demonstrates the “Conditional Mapping Logic engine (recruiter-validated rules)” line under AD2 above. The Calibre team’s ratified heuristics file is the proprietary domain layer that distinguishes this matching system from a generic LLM matcher.
Scope Change Process
If a feature request comes up that isn’t here:
- Map it to one of the 3 Annex 3 modules — does it fit?
- If yes → add to this doc with internal + audit-facing value notes
- If no → reject or defer to post-grant Phase 2
- Anything affecting cost, timeline, or grant defensibility goes to Calibre (Calibre) for sign-off
Annex 1.3 obligation: Calibre must inform EnterpriseSG in writing of any project variation. Material scope changes must go through Calibre before being added here.