Woven Talent · Internal

The hiring OS we will actually run.

Loom replaces TalentRecruit with one codebase: a shared RPO instance for many companies, and isolated single-company deploys. Recruiters add jobs. A job board is how candidates apply.

Feature checklist → Read the plan

Central RPO

One instance, many client companies. Jobs are owned by a company. Talent pool is instance-wide, with consent when a person is reused.

Isolated clients

Same image. Own Postgres, branding, WhatsApp, hostname. Exactly one company. No data crosses instances.

Job board

Internal dashboard to publish. Public listings to apply. Not a no-code career-site CMS in v1.

Locked plan

What we already decided

We are an RPO firm. One central instance holds many client companies. Isolated instances exist for clients who need their own DB and branding — those contain a single company.
Same application code for every deploy. Tenancy is config, not a fork.
Market: India-first, multi-country ready. DPDP now; GDPR-shaped fields from day one.
Jobs: Internal teams create and manage jobs in the dashboard. The Job Board dashboard publishes listings and powers the public apply flow.
AI: copilot, not autopilot. Parse, rank, draft, summarize. No ERIKA-style bot interviews in v1.

Instance model

flowchart LR
  subgraph code [Same Docker image]
    App[Loom app]
  end
  subgraph rpo [Central RPO instance]
    RpoDb[(Postgres: many companies)]
    RpoBoard[Job board: all clients or per-client]
  end
  subgraph iso [Isolated client instance]
    IsoDb[(Postgres: one company)]
    IsoBoard[Job board: that company only]
  end
  App --> rpo
  App --> iso

INSTANCE_MODE=rpo or single. The data model always has organizations. In single mode the switcher is hidden.

Interview · Aug 2026

Decisions, frozen

Change these in the repo when the product changes — not only in chat.

Business

Decision	Choice
What we are	RPO firm that also deploys isolated ATS instances for some clients
Central deploy	One shared instance, many companies inside it
Isolated deploys	One company per instance, own DB / branding / secrets
Code	Same application image for every instance
Market	India-first, then a few other countries (UK/SEA)
Compliance now	DPDP now; GDPR-shaped fields so we do not rework later

How work enters the system

Decision	Choice
Who creates jobs	Internal teams / recruiters, in the dashboard
Public intake	Job Board dashboard + public job board + apply pages
Not v1	TalentRecruit-style no-code multi-page career site builder
Not v1	One-click posting to Naukri / LinkedIn / Indeed (model so it can plug in)

Day-1 capabilities

Taken from what we actually depend on in TalentRecruit:

Job requisitions, approvals, JD templates / job library
Resume parsing, duplicate detection, talent pool search
Pipeline / stages / recruiter candidate tracking
Hiring manager portal (review, feedback, approvals)
Multi-client RPO: companies, job ownership, submissions to client HMs
Dashboards and reports
Campus / high-volume bulk hiring
Offer letters, e-sign, joining-doc checklist
Assessments / AI screening as copilot + knockouts + third-party tools
Interview scheduling + calendar + scorecards
Email + WhatsApp templates
Employee referrals and IJP

Explicitly not day-1: vendor portal, BGV, full career-site builder, job-board aggregator search.

AI stance. Copilot, not autopilot. Yes: parse, rank as a suggestion, draft messages, summarize interviews, generate JDs and scorecard questions. Not v1: conversational / voice bot interviews, autonomous stage movement.

Surfaces

Four products in one app

If a screen tries to be all four, it has failed. Recruiters run hiring from dashboards; candidates apply on a job board.

Person	Where they work	What they need
Recruiter	Recruiter OS	Speed: add jobs, publish, parse CVs, move pipeline, message, schedule
TA lead / RPO manager	Dashboards	Funnel, time-to-hire, recruiter load, client-company health
Hiring manager	HM portal	Review shortlists, score, approve. No sourcing UI
Employee	Light portal	Refer someone, apply to IJP
Candidate	Public job board	Find a role, apply, see status
Instance admin	Settings	Users, branding, pipelines, templates, WhatsApp/email, DPDP

Jobs and the job board

This is the spine. Recruiters create jobs. Publishing is a separate dashboard. Candidates never edit requisitions.

flowchart TD
  Recruiter[Recruiter] --> JobsDash[Jobs dashboard]
  JobsDash --> Draft[Draft / pending approval]
  Draft --> Approved[Approved job]
  Approved --> BoardDash[Job Board dashboard]
  BoardDash --> Public[Public listing + apply]
  Public --> App[Application]
  Recruiter --> Upload[Bulk CV / email ingest]
  Upload --> Pool[Talent pool]
  Pool --> App
  App --> Pipeline[Job pipeline]
  Pipeline --> HM[Hiring manager review]

Jobs dashboard

New job from blank, template, clone, or AI-assisted JD.
Company (required in RPO), department, location, employment type, openings, confidential flag.
India fields: CTC band, notice-period expectation, work mode.
Approval chain before the job is live internally.
Confidential jobs never hit the public board.

Job Board dashboard

Publish / pause / close. Preview, slug, SEO, card vs full page.
Apply form: resume, knockouts, custom fields per job.
Per-listing funnel: views → applies → screened → interview → offer.
Never bury Publish inside a 12-tab job edit form.

What this is not (v1)

A staffing ERP (billing, timesheets, contractor CRM).
An autonomous AI recruiter.
A replacement for Naukri as a sourcing UI.
An HRIS / 90-day onboarding suite.

Checklist

TalentRecruit, mapped

Every capability we researched. Recommendations match the August 2026 interview. Filter this list; accept or strike items in docs/FEATURES.md as the product evolves.

TalentRecruit sells an end-to-end suite (ERIKA + GAIA Autopilot). We do not clone that surface. We clone the jobs RPO recruiters and client HMs actually do, then add a job board.

1. Workforce planning and budgeting

Feature	Recommendation	Notes
Manpower plan / headcount budget vs hired	Later	Useful for enterprise TA; RPO can live without it in v1
Budgeted vs non-budgeted reqs	Later	Can be a flag on the job
Demand forecasting	Skip	Spreadsheet / BI until we have years of data

2. Job requisition management

Feature	Recommendation	Notes
Create job in recruiter dashboard	Keep v1	Primary intake for roles
JD templates / job library	Keep v1	You use this today
Clone / one-click duplicate	Keep v1	Campus + repeat RPO roles
Approval workflow	Keep v1
AI-assisted JD from a prompt	Keep v1.5	Copilot; recruiter edits
JD parse (paste a doc → fields)	Later	Nice, not blocking
Custom req forms per company / hiring type	Keep v1	Permanent vs contract vs intern
Confidential job flag	Keep v1	Hidden from board, search, unauthorized reports
Assign recruiters, HMs, panel	Keep v1
Configurable stages per job or company	Keep v1
Employment types (FTE, contract, intern, campus)	Keep v1

3. Career site vs job board

TalentRecruit’s Career Site Builder is a no-code branded microsite. Loom ships a job board instead.

Feature	Recommendation	Notes
Job Board dashboard (publish, pause, apply form, funnel)	Keep v1	Asked for explicitly
Public job list + job page + apply	Keep v1	Candidates enter here
Per-company branding (logo, color, domain)	Keep v1	Isolated instances especially
Combined RPO board + /{companySlug}	Keep v1	See Decisions
Knockout / screening questions on apply	Keep v1
Scheduled go-live for a posting	Later
No-code multi-page culture/CMS site	Later	TA should not be webmasters in v1
AI chatbot on the career site	Skip	Conflicts with copilot-not-autopilot
Multilingual career CMS	Later	Start English; add a language later

4. Job distribution (external boards)

TalentRecruit: one-click post to Naukri, LinkedIn, Indeed, and PortalRecruit unified search.

Feature	Recommendation	Notes
Shareable job link / WhatsApp / LinkedIn share	Keep v1	Free, high value
One-click Naukri / LinkedIn / Indeed post	Later	Model job_postings so this is an adapter
PortalRecruit-style unified board search	Later	Expensive; recruiters still use Naukri UI
Google for Jobs / IndexNow on our board	Keep v1	Cheap SEO win

5. Sourcing, referrals, IJP

Feature	Recommendation	Notes
Internal talent pool search (boolean + filters)	Keep v1
Employee referral portal	Keep v1.5	Day-1 need; can follow core pipeline
Referral tracking / bonus status field	Keep v1.5
IJP (employees apply to internal jobs)	Keep v1.5
AI auto-source from pool when a job opens	Later	Copilot suggested matches is enough
Internal mobility recommendations	Later

6. Candidate database (ATS core)

Feature	Recommendation	Notes
Resume parse (PDF/DOC/DOCX)	Keep v1
Duplicate detection + merge	Keep v1
Candidate 360	Keep v1
Pipeline / kanban + table	Keep v1
Bulk actions	Keep v1
Notes, tags, @mentions	Keep v1
Custom fields per org	Keep v1
Email-in / bulk CV upload	Keep v1	RPO + campus
Source tracking on every application	Keep v1
Do-not-contact / GDPR-DPDP hold	Keep v1

7. RPO / multi-client

TalentRecruit corporate is one-company; their agency SKU adds CRM. We are hybrid.

Feature	Recommendation	Notes
Many companies in one instance	Keep v1	RPO mode
Isolated one-company instance	Keep v1	Same code
Job owned by a company	Keep v1
Recruiter membership per company	Keep v1
Submission to client hiring managers	Keep v1	Unit of RPO work
Instance-wide talent pool with audit	Keep v1	Consent when reusing across clients
Client-facing HM login per company	Keep v1
Vendor portal (external agencies submit CVs)	Later	Not marked day-1
Client billing / invoice / margin	Skip	Not a staffing ERP

8. Engagement (email, WhatsApp, voice, chatbot)

Feature	Recommendation	Notes
Email templates + log on the person	Keep v1
WhatsApp templates + log	Keep v1 / v1.5	India-first; land in v1 if API access is ready
SMS	Later	WhatsApp covers India
In-app chatbot / voice bot screening	Skip	ERIKA territory
Automated drip sequences	Later	Start with stage-change templates
Unsubscribe / WhatsApp opt-out	Keep v1	Compliance

9. Screening, matching, assessments, AI interviews

TalentRecruit: ERIKA matching, bot interviews, Teams/Meet summaries.

Feature	Recommendation	Notes
Knockout questions	Keep v1
LLM rank / shortlist suggestion	Keep v1.5	Recruiter decides
Explainable match reasons	Keep v1.5	Trust
Third-party assessment links	Keep v1.5	HackerRank, Testlify, etc.
Homegrown question bank + proctoring	Skip
Conversational AI interviews	Later	Do not clone ERIKA in v1
Voice bot calls	Skip
Auto stage-move (GAIA Autopilot)	Skip	Dangerous in RPO
Interview question generator	Keep v1.5	Copilot
Transcript summary from a recording URL	Later	Meet/Teams bots are their own product

10. Interviews

Feature	Recommendation	Notes
Schedule interview, panel, type	Keep v1.5	Day-1 need; stack after pipeline if time-boxed
Scorecards	Keep v1	Needed with HM portal
Calendar OAuth (Google, Microsoft)	Keep v1.5
Candidate self-select from slots	Keep v1.5
Reminders	Keep v1.5
Video interview vendor embed	Later	Zoom/Meet links as fields first

11. Hiring manager portal

Feature	Recommendation	Notes
HM home: waiting on you	Keep v1
Review submitted candidates	Keep v1
Scorecard + comment	Keep v1
Approve interview / offer	Keep v1
No access to talent pool / job board admin	Keep v1	UX rule

12. Offers, e-sign, onboarding

Feature	Recommendation	Notes
Offer letter from template, approvals	Keep v1.5	After pipeline
Compensation fields (CTC)	Keep v1.5
Candidate accept / decline	Keep v1.5
Upload signed PDF / simple accept token	Keep v1.5
Digio / Zoho Sign	Later
Joining document checklist	Keep v1.5
Document library like TR onboarding	Later
First-90-days onboarding / HRIS	Skip	Different product
Offer benchmarking (market salary AI)	Skip

13. Background verification

Feature	Recommendation	Notes
Native BGV marketplace	Later	Not marked day-1
Status field + vendor link	Later	Cheap adapter later
Reusable verified identity	Later	TR differentiator; not ours yet

14. Campus / high volume

TR customers praise bulk CV handling for GET/campus drives.

Feature	Recommendation	Notes
Bulk CV upload	Keep v1
Drive / event object, college tag	Keep v1.5
Table-first pipeline (not only kanban)	Keep v1
Virtual assessment scheduling at scale	Later	Use assessment vendor
Registration microsite per drive	Later	Job board apply + source=campus may suffice

15. Analytics

TalentRecruit claims 200+ reports. We will not.

Feature	Recommendation	Notes
Funnel by job / company / recruiter	Keep v1
Time-to-hire, time-in-stage	Keep v1
Source of hire / apply	Keep v1
Recruiter load / SLA aging	Keep v1
Offer-to-join	Keep v1.5
Campus drive health	Keep v1.5
CSV export	Keep v1
200 canned reports	Skip
Predictive joining risk	Later

16. Compliance, admin, platform

Feature	Recommendation	Notes
Roles and permissions	Keep v1
Audit log	Keep v1
DPDP consent, retention, export/delete	Keep v1
Country privacy policy per org	Later	Start with instance + org text fields
SOC2/ISO program	Later	Process, not a feature flag
Custom fields, email templates, pipeline templates	Keep v1
API for partners	Later	tRPC first; REST when a vendor needs it
Mobile PWA for recruiters	Later	HM portal should already work on phones
Multi-currency / multi-country packages	Later	Fields exist; workflows later

17. TalentRecruit items we skip as identity

Feature	Recommendation	Notes
ERIKA as autonomous recruiter	Skip	Copilot only
GAIA Autopilot	Skip
Staffing billing, invoicing, lead CRM	Skip	RPO ATS, not agency ERP
Employee handbook / HRIS-in-ATS	Skip
Building our own assessment engine	Skip	Integrate

If v1 is slipping. Keep jobs + job board + parse + pipeline + HM review + RPO companies/submissions. Park WhatsApp, calendar, offers, referrals, campus drives, LLM rank. Never park: DPDP consent, duplicate merge, org isolation.

Craft

Best UX for the people who live here

Recruiters are primary. Density, keyboard, bulk actions, zero extra clicks to move a candidate.
Hiring managers get less — not a dumbed-down recruiter UI. Their job is decide.
Job board is a product, not a CSS theme on the ATS.
RPO chrome is optional. Company switcher only in rpo mode.
India-shaped forms. CTC and notice period are normal fields.

Recruiter OS

Home is work, not vanity KPIs: aging pipeline, interviews today, HM feedback waiting. Command palette everywhere (⌘K). Keyboard: c create, / search.

Hiring manager portal

Home is “Waiting on you.” Scorecard + one comment. Submit is the only primary button. Must work on a phone. Recruiters stay desktop-class in v1.

Manager dashboards

About twelve charts that answer real questions — time-to-hire, funnel, source, recruiter load, offer-to-join — with CSV on every table. Drill: chart → job → pipeline.

Sequence

v1, then close, then ecosystem

v1 — hiring OS + job board Auth, orgs, jobs dashboard, publish/apply, parse, pipeline, HM review, RPO submissions, email, basic analytics, DPDP, Docker.
v1.5 — communication and close WhatsApp, calendar, offers, referrals/IJP, campus drives, LLM copilot, assessment link-out.
v2 — ecosystem Naukri/LinkedIn/Indeed publish, vendor portal, BGV, Digio/Zoho Sign, recruiter PWA, optional control plane.

First coding slice

Org + auth → jobs dashboard → job board publish/apply → parse + pipeline + HM review. That loop is already an ATS.

Tenancy

Same image, many instances

flowchart TB
  subgraph image [loom Docker image]
    Web[Next.js + tRPC]
    Worker[BullMQ workers]
  end
  subgraph rpoInst [RPO instance]
    RpoPg[(Postgres)]
    RpoRedis[(Redis)]
  end
  subgraph clientInst [Isolated client]
    IsoPg[(Postgres)]
    IsoRedis[(Redis)]
  end
  image --> rpoInst
  image --> clientInst

Each instance owns database, Redis, object storage, branding, email, WhatsApp, and OAuth apps. A control plane that provisions instances is not v1.

Tenancy rules

Almost every business row has organization_id.
single mode seeds exactly one organization and never shows a switcher.
rpo mode: users see only companies they belong to, unless instance admin.
Candidates in RPO are people at instance scope; applications are per job.
No data crosses instances.

Core data model

Entity	Notes
organizations	Client companies. One row in single mode.
users	Auth identity.
memberships	user × org × role (recruiter, hm, employee, admin).
jobs	Requisition. organization_id, status, confidential, pipeline_id.
job_approvals	Chain of approve/reject.
job_postings	Job-board publication: slug, visibility, apply schema, published_at.
pipelines / stages	Template per org or per job.
people	Talent-pool person (parsed CV, CTC, notice, locations).
applications	person × job, current stage, source.
submissions	RPO: application sent to client HMs.
files	Resume versions, offer PDFs, joining docs.
messages	Email/WhatsApp thread items.
interviews	Time, panel, scorecards.
offers	Comp, approvals, accept/decline, joining date.
audit_events	Who did what, including cross-company reuse in RPO.
consents	DPDP purpose, captured at apply or parse.

Why this stack

One deploy for board and ATS

Layer	Choice	Why
App	Next.js App Router + TypeScript	Public job board needs SEO; recruiter app is authenticated. One deploy.
API	tRPC	End-to-end types; matches meridien.
DB	Postgres, one database per instance	Isolation you asked for.
ORM	Drizzle	Migrations we control.
Auth	Better Auth	Self-hosted, org-friendly.
UI	Tailwind + Radix primitives	Custom visual language — not a generic purple dashboard.
Tables	TanStack Table + virtualizer	Campus dumps of thousands of CVs.
Recruiter speed	CmdK command palette	High-volume RPO.
Queue	Redis + BullMQ	Parse, email, WhatsApp, ranking, PDFs.
Files	S3-compatible (R2 / AWS / MinIO)	Resumes and offer letters.
Email	SES or Resend per instance	Transactional templates.
WhatsApp	Meta Cloud API (Gupshup fallback)	India-first messaging.
Parse / rank	LLM → structured JSON + Postgres FTS	Affinda later if parse quality lags.
Calendar	Google + Microsoft OAuth	Self-serve slots in-app.
E-sign v1	Accept link + uploaded signed PDF	Digio / Zoho Sign later.
Observability	Sentry + structured logs	Per-instance DSN.
Package manager	pnpm	Same as other woven repos.

Explicit non-choices

Avoid	Reason
Mongo as source of truth	Hiring data is relational (job → application → interview → offer).
Prisma	Fine, but Drizzle is already in the Woven muscle memory.
Clerk / Auth0 as the only auth	Painful for many isolated instances and client SSO later.
Shared-schema multi-tenancy in one DB	Separate DBs per isolated client. RPO many-companies is inside one of those DBs.
Building a test-proctoring product	Integrate assessment vendors.
Kubernetes control plane on day one	Compose / one VM per instance until instance count hurts.

India-first data, not India-only code: CTC, notice period, serving notice, work mode are first-class columns.

Loom product spec · Woven Talent · sourced from the markdown in this repo · host the site/ folder anywhere static files work.