Chandra Jewels Pvt Ltd — Complete Business Blueprint & Product Roadmap

Section 1 — Executive Summary

Chandra Jewels Pvt Ltd is a custom jewellery manufacturer based in Seepz, Mumbai. It designs and manufactures gold, silver, and diamond-set jewellery for retailers and wholesalers, exporting 80–85% to the United States. The company processes 12–15 new designs and 5–7 repeat orders daily across 13 active export customers, with all coordination currently running through WhatsApp groups, Excel sheets, and institutional memory held by a small number of key people.

The business has proven product quality, strong customer loyalty, and a repeatable manufacturing process. Its single largest growth constraint is operational: the founder (Anmol Jain) sits in too many critical paths — every pricing decision, every design review, every production escalation. The company cannot scale beyond its current volume without systematising these paths.

Chandra OS is the internal operating platform that replaces WhatsApp-as-workflow with a structured, event-driven, AI-assisted system — issuing a unique Job ID at inquiry intake and tracking every piece from brief to dispatch. It does not replace Gati (ERP) or skilled human judgment in design and manufacturing. It replaces coordination overhead, manual status tracking, and memory-dependent approvals.

12-month outcome: Anmol spends ≤2 focused hours/day on operations (exceptions and product development only). Every department has a real-time queue. Delays are predicted 7–14 days before they happen. The founder's knowledge is encoded in checklists and pricing rules, not retrieved from his head on demand.

Section 2 — Business Overview

2.1 What We Are

Chandra Jewels Pvt Ltd is a B2B jewellery manufacturer — not a retailer. Customers are jewellery retailers, wholesalers, and brands globally. We manufacture from scratch: from design concept through CAD, casting, stone setting, polishing, and dispatch.

2.2 Business Units

The factory is the core unit. This blueprint focuses on the factory (B2B manufacturing).

2.3 Geography & Markets

• Manufacturing: Seepz, Mumbai (factory) + Domjur/Calcutta (CAD operations)
• Primary export: USA — 80–85% of total revenue
• Secondary markets: Hong Kong, UK, UAE

2.4 Ownership Structure

Five partners total: 3 founding partners (Anmol Jain, Chirag + 1), 2 new investors who acquired equity in the last month.

2.5 Current Scale

Section 3 — Products & Manufacturing

3.1 Jewellery Categories

3.2 Metals & Stones

Metals: Gold (10K, 14K, 18K — yellow, white, rose), Sterling silver, Platinum, Various alloys

Stones: Natural diamonds (VVS, VS, SI), Lab-grown / CVD diamonds, Coloured gemstones and moissanite, Fancy cuts: baguette, princess, cushion, mixed shapes

Findings (outsourced): Lobster clasps, jump rings, GS locks, chains.

3.3 Design Process: Coral vs. CAD

Section 4 — Full Manufacturing Journey

Step 1 — Inquiry Intake

Customer sends inquiry via WhatsApp group or email. Content: reference image, design description, budget, diamond specs, size, delivery requirement. Assigned coordinator (salesperson/RM) receives, reads, and forwards to appropriate design team. Anmol reviews every design in a one-page format before it goes to the customer. Conversion rate highest when design is sent within 24 hours of inquiry.

Step 2 — Coral Design (2D)

Coral designer creates 2D layout per instructions. Salesperson checks, prices, creates one-page summary. Anmol reviews and approves before sending to customer. Customer approves on WhatsApp or email (or requests changes → loop).

Step 3 — CAD Design (3D)

Once Coral is approved (or for direct CAD jobs), CAD designer creates full 3D model. Internal QC: weight checked against budget, dimensions checked against instructions, Coral match verified. Anupam (CAD Ops Head) reviews. If Coral was skipped: CAD sent directly to customer for approval.

Step 4 — Pricing & Quotation

Anmol's reverse-engineering method (currently manual): Take customer target budget → Deduct duties (~20%) → Split: ~50% gold cost, ~50% diamond cost → Gold cost ÷ current gold rate = gold weight → Diamond budget ÷ client's fixed per-carat rate = diamond carats → Add labour → Add wastage → Verify margin above client floor. Labour rates, diamond per-carat rates, and margins are fixed per client at onboarding.

Step 5 — Customer Approval

Coral or CAD image + price sent to customer group. Aayushi tracks all pending approvals in a Google Sheet. Three outcomes: Approved / Changes requested / Rejected. On approval: forwarded to Proof Group, logged in approval sheet.

Step 6 — Style Master Creation (Gati)

Style master created in Gati ERP by Calcutta team. Style master captures: karat, diamond sizes, ring size/bracelet length, findings, chain type, all production specs. First point at which a unique style number exists. Order placed in Gati — triggers diamond allocation check. Currently: Anmol reviews unprocessed order backlog (can reach 10–15 days).

Step 7 — CAM (Wax Printing)

CAD file exported to CAM format → loaded into wax printing machine. Machine prints 3D physical wax model of the piece. Alternative: pull wax from rubber mould (for high-volume repeat pieces). Multiple wax pieces assembled into a wax tree for batch casting.

Step 8 — Casting

Wax tree placed in metal flask → Flask filled with investment powder → Flask goes into furnace — wax burns out → Molten gold poured into void → Flask cooled → Raw gold castings cut from tree.

Step 9 — Filing

Remove casting roughness and sprues. Manual skilled work — surface prep before polish.

Step 10 — Pre-Polish

Polish metal surface, especially under diamond seats. Done before setting so the under-seat area is clean.

Step 11 — Diamond Setting

Always manual, always human — no machine can set diamonds. Skilled setters place each stone per the CAD layout.

Step 12 — Linking

Assembly of multi-component pieces. Sequence relative to setting is piece-dependent.

Step 13 — Final Polish

Full surface polish after all setting and linking is complete.

Step 14 — Final QC

Check against every customer instruction recorded at approval stage. CVD test: machine test to detect lab-grown diamonds in natural diamond orders (critical). Every piece, every time.

Step 15 — Rhodium / Plating

White gold: rhodium plating is mandatory. Yellow/rose gold: plating per customer specification.

Step 16 — Post-Rhodium QC → Packing → Dispatch

Re-QC after plating. Pieces packed per customer, per invoice. Standard shipping days: Monday, Tuesday, Saturday. Express possible for urgent orders.

Section 5 — People & Organisation

Note: Client managers are coordinators and relationship managers, not traditional salespeople. Sales is driven by Anmol (design quality) and Chirag (in-person relationship building).

Section 6 — Current State: What Works & What Doesn't

What Works

• Quality: The manufacturing process is best-in-class and already invested in.
• Design speed: 24–48 hour design turnaround is genuinely differentiated.
• Pricing discipline: Fixed per-client labour and diamond rate grids.
• Customer loyalty: Long-term customers trust the quality.
• Repeat business: 130–200 repeat orders/month shows strong retention.
• CAD team structure: Division by category works well operationally.

What Doesn't Work

Section 7 — USP & Competitive Edge

Chandra's competitive advantage is not the machines. It is:

1. Design quality encoded in institutional knowledge: After years of working with each customer, Anmol and senior designers know exactly what each customer prefers. A $10,000 pendant brief is interpreted correctly on the first attempt.
2. 24–48 hour design turnaround: The psychology of the buyer is hottest on the day of inquiry. Faster design → higher conversion → more orders.
3. Right product the first time: Correct weight, correct stones, correct finish — delivered without multiple revision cycles.
4. Delivery commitment: Occasion-based delivery is treated as a hard deadline. The factory has demonstrated 3–4 day full manufacturing cycles when required.
5. Trust built over years: Customers transfer work from other factories to Chandra because of reliability.

The product goal is to encode points 1–4 into systems so they scale beyond Anmol's personal bandwidth.

Section 8 — The Product Vision: Chandra OS

What It Is

An internal operating platform that wraps all coordination from customer inquiry to shipment dispatch. Every inquiry becomes a structured Job. Every stage has a queue. Every SLA has an automated escalation.

What It Is Not

• Not a replacement for Gati (ERP stays as system of record)
• Not a customer-facing storefront
• Not a system that removes skilled human judgment from design or manufacturing
• Not a new WhatsApp — customers keep using WhatsApp; it just becomes an input channel

Three Architectural Principles

1. Job-centric, not message-centric: Today the unit of work is a WhatsApp message. In the future state, the unit of work is a Job with ID CJ-2026-000123. WhatsApp becomes an input channel only.
2. Event-driven, not poll-driven: No human asks "what's pending?" The system emits events (cad.submitted, approval.received, sla.breached) and queues react automatically.
3. Queues, not inboxes: Every role has a prioritised work queue. AI ranks the queue. No one decides "what should I work on next?" — the system tells them.

Section 9 — Core Object: The Job

Everything orbits around one record — the Job.

job
├── job_id              CJ-YYYY-NNNNNN  (issued at intake)
├── client_id           → clients table
├── client_manager_id   → users (Rupsa / Rumi / Trishna)
├── category            bridal | hiphop | cuban | gemstone | custom
├── sub_category        pendant | ring | chain | bracelet | earring | bangle
├── parent_style_id     → styles (if repeat or variant)
├── intake_channel      whatsapp | email | call | portal
├── brief_structured    JSONB (AI-extracted from intake message)
├── files               JSONB[] (refs, CAD, renders, CAM, photos)
├── priority_score      float 0–100 (continuously recomputed by AI)
├── delivery_promise    date
├── current_state       enum (25 states)
├── current_owner_id    → users
├── sla_targets         JSONB per-stage TAT
├── sla_breach_count    int
├── value_estimate      decimal
├── diamond_required    JSONB (synced from Gati)
├── gati_order_id       string
└── risk_score          float 0–100 (AI-predicted delay risk)

Supporting objects: job_event (append-only ledger), job_file (every design file with version and hash), job_revision (every change request), job_communication (every WhatsApp/email message).

"The job_event ledger is what kills 'Aayushi as memory' — the database remembers everything."

Section 10 — Job State Machine (25 States)

DRAFT → INTAKE_PARSED → BRIEF_CONFIRMED
  ↓
COREL_QUEUED → COREL_IN_PROGRESS → COREL_SUBMITTED
  ↓
CAD_QUEUED → CAD_IN_PROGRESS → CAD_INTERNAL_QC → CAD_SUBMITTED
  ↓
CLIENT_APPROVAL_PENDING ⇄ REVISION_REQUESTED → (loop to CAD)
  ↓
CAD_APPROVED
  ↓
PRICING_QUEUED → QUOTE_SENT → QUOTE_APPROVED
  ↓
PO_RECEIVED → STYLE_MASTER_CREATED → GATI_ORDER_CREATED
  ↓
CAM_QUEUED → CAM_READY
  ↓
CASTING → FILING → SETTING → POLISHING → FINAL_QC
  ↓
DISPATCH_READY → SHIPPED → DELIVERED → CLOSED

Orthogonal states (apply to any stage): ON_HOLD, CANCELLED, ESCALATED, AT_RISK

Every transition has: Guards (preconditions), Actions (side effects), SLA (time allowed).

Example transition:

CAD_SUBMITTED → CLIENT_APPROVAL_PENDING
  Guards:   cad_file_present, render_present, internal_qc_passed, weight_within_tolerance
  Actions:  notify_client_via_approved_channel, start_sla_timer(72h), log_event
  SLA:      72h
  Breach:   escalate(Client Manager → Aayushi → Founder)

Section 11 — Queue Architecture

One prioritised queue per work role. AI ranks every queue.

Priority Score Formula

priority = (0.30 × urgency)          # days_to_promise, inverted
         + (0.20 × client_tier)      # Tier A=1.0, B=0.7, C=0.4
         + (0.15 × value)            # log-scaled order value
         + (0.15 × downstream_risk)  # how many other jobs this blocks
         + (0.10 × aging_in_stage)
         + (0.10 × revision_pressure)
         - penalty_if_blocked

Section 12 — AI Layer — Eight Engines

12.1 AI Intake Engine

Input: WhatsApp message (text + images + voice), email, call transcript. Output: Structured brief_structured JSON. Stack: Claude / GPT-4o + jewellery-domain classifier + CLIP image embeddings + internal glossary. Key behaviour: Ambiguities trigger an AI clarification bot before a human sees the message. 80% of briefs should parse without human intervention.

12.2 AI Routing Engine

Decides which queue a job enters based on category × sub-category × client preferences × designer skills × current workload. Skips Coral for repeat styles and routes directly to CAD.

12.3 AI Priority Engine

Recomputes priority score on every event. Surfaces the top-N jobs per worker. Automatically re-ranks when delivery dates change, diamonds go short, or revisions are received.

12.4 AI Revision Intelligence

Diffs CAD versions and clusters revision causes. Flags clients and designers with abnormal revision rates. Predicts revision probability: "this client revised 7 of last 10 pendants on bail thickness — alert designer before submission."

12.5 AI Shipment Predictor

Monte Carlo simulation over historical stage TATs. Outputs P50/P80/P95 ship dates and a risk score. When risk > 0.7, the job auto-escalates before it slips. This is the biggest single operational win: failure prediction, not failure post-mortem.

12.6 AI Workload Balancer

Continuously watches queue depth × worker capacity × skill match. Auto-reassigns or flags overload. Prevents all urgent work piling onto one designer.

12.7 AI Customer Communication Engine

Generates status updates per client communication preferences. Drafts go to the client manager for one-click send, or auto-send for trusted templates. Handles 80% of "any update?" follow-ups without human writing.

12.8 AI Risk & Delay Detector

Patterns that trigger risk.detected: Stage age > SLA threshold, Diamond shortage probability above threshold, Quote not approved within X days, CAD revision loop > 3 cycles, Client approval not received and delivery date < N days away.

Section 13 — Module Specifications

13.1 Intake & WhatsApp Ingestion

WhatsApp Cloud API integration, paste-from-WhatsApp captures full thread, voice-note auto-transcription (Whisper API — handles Hindi/English code-switching), reference images uploaded to S3, mandatory structured fields enforced, AI clarification bot handles ambiguous briefs.

13.2 Pricing Module

Inputs: Metal type + karat, weight, diamond data, labour category, finish type. Rate Sources: daily-locked gold rate, per-client diamond rate matrix, per-category labour, wastage %, margin with configurable floor. Rules: every line item shown separately, every recalculation creates new version, revision shows "what changed" summary.

Anmol's reverse calculation, encoded:

net_budget = customer_budget × (1 - duty_rate)
gold_budget = net_budget × metal_split_ratio
diamond_budget = net_budget × stone_split_ratio
gold_weight_g = gold_budget / current_gold_rate[karat]
diamond_carats = diamond_budget / client_diamond_rate[size][clarity][color]
labour = gold_weight_g × client_labour_rate[category]
total = gold_budget + diamond_budget + labour + wastage + margin

13.3 Approval Module

Customer receives a link — sees rendered CAD + price + spec sheet → taps Approve / Request Changes / Reject with comments. WhatsApp carries the notification; the decision lives in the system, timestamped.

13.4 PO Check & Style Master Module

Three-way match before production: Approved enquiry spec + Approved CAD spec + Gati entry. Style master pre-fill: AI pre-fills 90% of Gati fields. Auto-generated style numbers. Spec lock: Once quotation is approved, spec is frozen.

13.5 Production Planning Module

Every production stage is a queue with a due date. AI back-calculates required start dates. Late-stage detector flags missed windows immediately. Physical screens in factory show department-specific dashboards.

13.6 Diamond Inventory Module

Complete diamond inventory in system. When order placed, required diamonds immediately reserved. If insufficient, purchase order raised automatically — not discovered at casting. 7/14/30-day shortage forecast.

13.7 Repeat-Order Path

Customer says "Make like CJ-2026-0847 in 18kt instead of 14kt" → App pulls original enquiry + CAD + pricing → Auto-generates new Job ID → Skips Coral/CAD if design unchanged → One-click quotation if delta within tolerance. Target: cut repeat cycle time from ~5 days to <1 day.

13.8 Dashboards

• Founder/CEO: Shipment risk heatmap, Top-10 high-value jobs, L3 escalations, Revenue pipeline by stage, Capacity vs demand
• Anupam (CAD Ops): Queue depth per designer, revision loop alerts, internal QC failures
• Aayushi: Approval aging, SLA breach list, exception queue
• Yogi (Production): Department WIP, casting calendar, diamond-readiness, at-risk ship dates
• Client Managers: Their clients only — pending approvals, pricing queue, dispatches, comm log
• Diamond Dept: Required vs. issued vs. shortage forecast
• Client Portal (external): Their jobs, renders, approvals, status, invoices, tracking

Section 14 — Integration Architecture

┌─────────────────────────────────────────────────────────────┐
│ L7  PRESENTATION    Founder · Dept · Client · Mobile PWA    │
├─────────────────────────────────────────────────────────────┤
│ L6  AI INTELLIGENCE Intake · Routing · Priority · Risk ·    │
│                     Revision · Shipment · Workload · Comm   │
├─────────────────────────────────────────────────────────────┤
│ L5  ORCHESTRATION   Workflow Engine (Temporal)              │
│                     State Machines · SLA Engine · Escalation│
├─────────────────────────────────────────────────────────────┤
│ L4  SERVICES        Job · Design · CAD · Approval · Pricing ·│
│                     CAM · Production · Diamond · Shipment   │
├─────────────────────────────────────────────────────────────┤
│ L3  EVENT BUS       Redpanda/Kafka — all state changes emit  │
├─────────────────────────────────────────────────────────────┤
│ L2  DATA            Postgres · S3 (files) · OpenSearch       │
│                     Vector DB (design refs) · ClickHouse     │
├─────────────────────────────────────────────────────────────┤
│ L1  INTEGRATION     WhatsApp Cloud API · Gati ERP · Email ·  │
│                     Rhino/Matrix CAD · Courier APIs          │
└─────────────────────────────────────────────────────────────┘

Gati Bridge

Chandra OS wraps Gati — does not replace it. Chandra OS → Gati: style master create, BOM, order create, dispatch trigger. Gati → Chandra OS: diamond stock, allocation, inventory, invoice numbers, dispatch confirmation. If Gati exposes API: direct. If not: RPA layer (Playwright) + nightly reconciliation.

CAD/CAM Plugin

One-click submit from CAD workstation. Eliminates manual file forwarding via WhatsApp. CAD versioning: every save creates v_n, diff viewer for revisions. AI manufacturability check: wall thickness < 0.8mm, prong issues, undercuts, weight vs. target.

File Storage

S3 + metadata in Postgres + OpenSearch + vector embeddings. Every file tagged: job_id, client_id, style_id, category, version, kind, hash. All access via short-expiry signed URLs (15 min).

Section 15 — Tech Stack

Build vs. buy rule: Temporal, Redpanda, Postgres, ClickHouse, OpenSearch — buy or use OSS. Build only: Job logic, pricing engine, AI engines, dashboards.

Section 16 — Implementation Roadmap — 4 Phases / 12 Months

Phase 1 — Foundation (Months 1–3)

Goal: Kill WhatsApp as workflow. Issue a Job ID at intake.

Pilot: One client (Jewelry Unlimited) + one category (Hip Hop pendants). Run parallel with WhatsApp for safety.

Phase 2 — Approval, Pricing, ERP Bridge (Months 4–6)

Goal: Automate approval tracking, encode pricing rules, connect to Gati.

Roll out to: All Hip Hop + Cuban clients.

Phase 3 — Production & Diamond Orchestration (Months 7–9)

Goal: Full production visibility. Diamond inventory live.

Roll out to: Bridal + all remaining clients.

Phase 4 — Full AI Layer & Optimisation (Months 10–12)

Goal: Anmol ≤2 hours/day. AI handles 80% of routine coordination.

Section 17 — Hiring Roadmap

Section 18 — KPI Framework

Operational KPIs

Financial KPIs

AI / Automation KPIs

Section 19 — The Three Non-Negotiables

1. Single Job ID from intake to dispatch. No parallel tracking. Enforce ruthlessly from Phase 1.

2. Event ledger is sacred. Every state change is an event. No silent updates. This is what makes the organisation auditable, the AI trainable, and disputes resolvable on data.

3. Humans own creative and exception work only. If a human is doing coordination — forwarding messages, chasing status, manually re-entering data — it is a system gap. File it as a bug. Fix it in the next sprint.

What This Removes From Humans

Part B — Detailed Feature Specifications

How to Use: Each feature has a unique Spec ID. To generate code for one feature, share only its spec file with the developer — it is self-contained.

ID Format: COS-[MODULE]-[NNN]

Module Codes

Spec Status Tags: DRAFT   READY   IN DEV   DONE

Spec Index

More specs added as each module is detailed.