B2B Marketplace and Product Platform

Indian Specialty Coffee Sourcing Platform

A coffee marketplace system that treats export readiness, sample logistics, and role-aware workflow state as part of the product core, not post-conversion cleanup.

Trust before conversion Marketplace product

Context

The product is built around a simple marketplace insight: discovery is not the hard part. Conversion quality is. In specialty coffee, buyers need more than origin stories and tasting notes. They need to know whether a supplier is commercially ready, whether samples can be shipped, which certifications are real, and where the lot sits in the inquiry pipeline. That pushes the product beyond catalog UX into workflow design, trust modeling, and role-aware system behavior.

Problem

Cross-border coffee sourcing breaks down when buyers can discover lots but still cannot tell whether a supplier is actually ready to transact. Discovery without readiness, sample logistics, compliance context, and clear next-step workflow creates weak leads, low-confidence matching, and long inquiry cycles.

What I Built

A supplier onboarding system that captures export capability, sample readiness, documentation status, certifications, and commercial readiness as structured fields
A coffee-lot catalog with readiness-aware filters for region, process, varietal, certifications, export status, and sample availability
Buyer preference capture and explainable ranking logic that scores lots against profile fit, readiness, and operational viability
Role-gated buyer, supplier, and admin surfaces over a shared marketplace data model with explicit permissions and workflow state
Inquiry and sample-request flows that keep lot state, buyer intent, and supplier response history visible instead of scattering them across email threads

Notes

Overview

This project is a B2B marketplace system for sourcing Indian specialty coffee, not just a listing site. The core idea is that buyers should be able to browse lots with enough structured context to act confidently: export readiness, sample logistics, compliance responsibility, documentation state, and fit to buyer preferences.

System shape

The product works as a multi-role marketplace with a shared operating model underneath:

Supplier onboarding -> lot creation -> readiness validation -> buyer discovery ->
recommendation ranking -> inquiry / sample request -> admin review -> conversion tracking

That matters because coffee sourcing is not a single-page search problem. It is a chain of decisions, validations, and handoffs. If the system does not preserve that state clearly, the workflow leaks back into email threads and spreadsheets.

What the product covers

supplier onboarding for export, sample, and documentation readiness
searchable lot catalog with readiness-aware filters
role-gated buyer, supplier, and admin workspaces
sample-request workflow with live inquiry tracking
explainable recommendation scoring tied to buyer preferences and lot readiness
review capture and interaction logging for later ranking improvements

Core entities and data model

The platform becomes much more useful once the marketplace entities are explicit:

suppliers: exporter identity, certifications, ship-to markets, sample capability, and documentation readiness
coffee_lots: origin, farm or estate, varietal, process, harvest information, cupping notes, inventory posture, and commercial metadata
buyer_profiles: roast style, flavor preferences, certifications required, target volumes, and logistics preferences
sample_requests: request state, quantity, shipping status, buyer intent, and supplier response history
recommendation_events: scored matches, explanation payloads, and interaction feedback
admin_reviews: exceptions, verification checks, and readiness overrides when the platform needs a human gate

That schema turns the marketplace from a passive catalog into a system that can reason about fit, readiness, and workflow state.

Recommendation and ranking logic

The first version of the ranking layer should stay explainable.

A useful score can combine:

buyer preference fit
lot quality and certification fit
export readiness
sample availability
destination compatibility
supplier responsiveness

Instead of a black-box rank, the UI should be able to say something like:

Ranked highly because this lot matches washed-process preferences, meets export-readiness requirements,
supports sample shipment, and aligns with the buyer's target flavor profile.

That explanation matters. In B2B matching, trust comes from visibility into why the system thinks a lead is worth pursuing.

Workflow state and operational visibility

The platform also needs explicit state transitions for inquiry handling:

Draft inquiry -> submitted -> supplier acknowledged -> sample requested ->
sample shipped -> buyer reviewed -> negotiation -> closed / dropped

Without that, the product becomes a thin handoff into manual follow-up. With it, the platform can answer important questions:

where deals stall
which suppliers respond quickly
which lot attributes correlate with real follow-through
whether recommendation quality is improving over time

Why it stands out

The strongest part of the concept is that the trust layer is inside the marketplace itself. Instead of assuming a buyer will sort out logistics after discovery, the product makes readiness and responsibility visible upstream in the browsing and ranking experience.

That shifts the platform from “coffee listings” to a product where data modeling directly improves matching quality and conversion confidence.

Product and technical depth

What makes this feel like a full system rather than a concept page is the way each layer supports the next:

onboarding produces trusted supplier state
trusted supplier state improves lot ranking
ranking quality improves inquiry quality
inquiry events create better training data for future ranking
admin review closes the loop when marketplace logic needs human judgment

That is the part I find most interesting. The product is nominally about coffee, but the engineering problem is really about marketplace trust, workflow state, recommendation transparency, and turning operational constraints into product logic.

Role / ownership

Owned the product framing around trust-before-conversion and translated that into a concrete marketplace data model
Designed the system around supplier onboarding, lot-readiness modeling, recommendation transparency, and inquiry workflow state
Focused on making the platform behave like a real B2B operating system rather than a static directory of coffee listings

Impact

Moves coffee sourcing beyond listing pages by turning logistics, compliance, and supplier readiness into first-class product inputs that improve discovery quality, recommendation trust, and inquiry conversion.

Stack

FastAPI
Python
Postgres
Supabase Auth
Role-based access
Recommendation scoring
Workflow state
Marketplace data modeling

Technical design

Supplier onboarding model that captures export capability, sample-shipping readiness, compliance ownership, and documentation status as first-class structured data
Marketplace schema with suppliers, lots, certifications, buyer profiles, sample requests, recommendations, inquiries, and audit events
Coffee-lot catalog with search, filtering, ranking, and explanation surfaces backed by a shared lot-readiness model
Buyer preference and inquiry workflow tied to live lot data, recommendation scoring, and downstream state transitions
Role-based surfaces for buyers, suppliers, and admins backed by shared service logic and row-level authorization boundaries

Engineering decisions

Treat export, sample, and compliance fields as approval gates rather than optional notes
Keep the first recommendation layer rules-based and explainable so users can trust why a lot was ranked
Separate buyer, supplier, and admin surfaces with role-based access instead of forcing one overloaded interface
Log lot views, recommendation clicks, reviews, and inquiry states so the platform can later shift from heuristic ranking to behavior-informed ranking
Model inquiry and sample workflows as explicit states instead of free-form CRM notes so the platform can reason about conversion health

Tradeoffs

Structured onboarding improves trust, but it raises supplier setup friction early on
Explainable recommendations are easier to audit, though they can feel less magical than a black-box ranking model
Deeper readiness and logistics modeling makes the product more defensible, but it pulls scope beyond a simple listing marketplace
Role-specific surfaces improve usability, though they require more careful schema and permission design than a single shared dashboard

Outcome / impact

Buyers can evaluate whether a lot is commercially viable before starting a sample request
The ranking layer can favor commercially realistic lots instead of just attractive listings
The product establishes a stronger moat around data quality, matching logic, and conversion workflow instead of listing volume alone
Supplier and admin workflows stay closer to the underlying system state instead of drifting into disconnected manual coordination

Lessons learned

In a B2B marketplace, the most important product work often happens before the lead is created
Recommendation systems become more credible when readiness and operational viability are treated as first-class ranking inputs
Role-aware systems need a shared source of truth underneath them, or marketplace state quickly fragments into manual follow-up