Frenzee
Methodology + supplier index
Methodology + supplier index — behind the live demo

How Frenzee goes from "I want to make X" → 3 ranked matches.

Everything below is the methodology behind the demo — how the supplier graph was built, what the framework filters on, how matches are ranked. The full ~19K-record dataset (Alibaba + IndiaMART) is browsable at the bottom, plus a country breakdown. Click any row to verify against the source.

← Back to live demo Alibaba + IndiaMART · ~19K supplier listings ~12.5K unique companies · 39 countries
How it works

Brand brief + ten supplier filters → ranked matches

The framework starts one step upstream of sourcing. Dimension 0 is a brand brief — product-consumer fit, captured as free text — that grounds every later spec choice. Then Dimensions 1–10 are the buyer-intent filters, each mapped 1:1 to a supplier-schema field. Click any dimension to expand.

1
Buyer intent
"I want to make 200 oversized cotton tees, 240+ GSM, ivory + sage, $4 landed, US-bound."
→
2
Brand brief + 10-dim extraction
Dim 0 captures buyer + need + occasion. Dims 1–10 capture product, fabric, qty, channel, geography, timeline, compliance, customization, samples, trust.
→
3
Filter against supplier graph
Each dimension maps to a supplier-schema field. AND/OR logic returns a candidate set, ranked by fit score.
→
4
Top 3 ranked matches
Trade-off notes, trust signals, first-message preview ready to fire.

The 11 dimensions — click to expand

Q0
Brand brief — product-consumer fit (does not filter suppliers — grounds Dims 1–10)
→ free-text brief · captured once, played back when buyer drifts
▾

The brief before the brief. James Cunningham, brand-build advisor: "It's a product-consumer match. It doesn't change the supplier-brand owner matching — it's the step before it in the value chain."

Two ways in (run whichever applies)

  • Brand-first: You've got Brand X. Have you defined the right product around the occasion they'd wear it for?
  • Consumer-first: You're serving Consumer X. Have you defined the right product around a proven need?

The 6 sub-prompts

  • 1. Intended buyer — gender
  • 2. Intended buyer — age
  • 3. Climate / vibe of their targeted consumers' market
  • 4. Stage in life
  • 5. Functionality wanted — the need they're looking to sort out by using this product
  • 6. Why are they buying this from you specifically?

What Frenzee does with it

  • Drift detector — every later spec change cross-checked against the brief. "You said warm-climate Gen Z streetwear, but you're spec'ing 240gsm — that reads light-tee, not heavyweight hoodie."
  • Marketing payload — the "why from you specifically" line is the seed for the launch tagline + first ad copy. Played back when the brand goes live.
  • Coherent supplier message — the brief informs the first message to the shortlisted suppliers, so context lands in one shot instead of a back-and-forth.
Q1
Product Type — what are you making?
→ supplier.top_products · category + sub-type match
▾

10 major categories, ~150 sub-types. Drives every downstream filter.

Tops

  • T-shirts: crew, V, scoop, henley, ringer, pocket, longline, raglan, regular, slim, oversized, boxy, fitted, ribbed
  • Polos: pique, jersey, performance, long-sleeve
  • Button-downs: oxford, dress, chambray, flannel, tunic, popover, camp-collar, western
  • Blouses: peasant, peplum, wrap, off-shoulder, smock, cami, blouson, ruffle
  • Tanks: classic, racerback, halter, tube, ribbed, cropped, athletic
  • Sweaters: crew, V, turtleneck, cardigan, cable, fisherman, oversized, cropped
  • Hoodies / sweatshirts: pullover, zip-up, cropped, oversized, half-zip

Bottoms

  • Pants: jean, chino, trouser, dress pant, cargo, jogger, track, sweatpant, paperbag, palazzo, wide-leg, straight, skinny, bootcut, flare, mom, boyfriend, tapered
  • Skirts: mini, midi, maxi, A-line, pencil, pleated, wrap, denim, leather, tiered, bias-cut
  • Shorts: bermuda, biker, board, running, denim, dress, paperbag, cargo, athletic, hot pants, culottes
  • Leggings: classic, high-waist, biker, capri, fleece-lined, ribbed, seamless, performance
  • Jumpsuits / rompers

Dresses + Outerwear

  • Dresses by length: mini, midi, maxi, gown
  • Dresses by silhouette: shift, sheath, A-line, fit-and-flare, wrap, slip, bodycon, bandage, smock, peplum, mermaid
  • Coats: trench, peacoat, parka, puffer, duffel, cape, wool, down, wrap
  • Jackets: blazer, bomber, denim, leather, varsity, biker, utility, anorak, windbreaker, fleece, quilted, shacket, harrington, chore

Lingerie / sleepwear / underwear

  • Bras: balconette, bralette, push-up, sports, plunge, T-shirt, wireless, longline, strapless
  • Panties: brief, bikini, thong, hipster, boyshort, high-waist, seamless, G-string, tanga
  • Sleepwear: matching pajama set, separates, nightgown, chemise, robe, kimono
  • Hosiery / socks: tights, stockings, ankle, crew, knee-high, athletic, performance

Activewear · Accessories · Specialty

  • Activewear: sports bras (light/medium/high impact), leggings, biker shorts, joggers, tank, half-zip, jacket
  • Bags: tote, crossbody, backpack, clutch, satchel, duffel, weekender, fanny pack, bucket, sling
  • Hats: baseball, bucket, beanie, fedora, sun, beret, visor
  • Footwear (separate cluster): sneakers, dress shoes, loafers, boots, sandals, flats, heels, mules
  • Children's wear (CPSIA-required): all + onesies, sleepwear, school uniforms
  • Specialty: modesty / hijab · uniforms · bridal / formalwear · maternity · adaptive
Q2
Material — look + feel descriptors
→ supplier.fabric_strengths · layman→industry translate
▾

The translation engine. The buyer says "soft and breathable for summer" — the AI maps to cotton, linen, modal, bamboo viscose and intersects with each supplier's fabric strengths.

Sample of the 60+ row translation table

  • Soft, breathable, summery → cotton lawn, voile, linen, modal, bamboo viscose, rayon
  • Drapey, slip-feel → silk charmeuse, cupro, viscose satin, modal jersey
  • Crisp, structured → cotton poplin, twill, oxford, ponte, taffeta
  • Stretchy, body-hugging → jersey + 5% elastane, ponte, scuba, performance interlock
  • Warm, cozy, winter → wool melton, cashmere, fleece, French terry brushed, sherpa
  • Heavyweight tee → 220–300 GSM jersey, ringspun cotton, Supima
  • Sweater-like, no itch → cotton knit, merino wool, bamboo blend, viscose-nylon knit
  • Denim — soft, broken-in → 9–11 oz right-hand twill, stretch denim 1–2% elastane
  • Activewear / wicking → nylon-spandex (Supplex), poly-spandex interlock, brushed poly knit
  • Lingerie / bra fabric → stretch lace, mesh tulle, microfiber jersey, power mesh
  • Sheer / chiffon → polyester chiffon, silk georgette, mesh, organza
  • Faux leather jacket → PU on knit/woven (drape), PVC on woven (cheaper)
  • Eco / sustainable → GOTS organic cotton, recycled poly (GRS), Tencel/Lyocell
  • Velvet-looking → velvet, velour, plush, chenille
  • Compression / shapewear → high-elastane jersey (15–25%), power mesh

Naturals

  • Cotton (Pima, Supima, Egyptian, organic, BCI, ringspun, combed)
  • Linen, hemp, ramie
  • Silk (mulberry, charmeuse, chiffon, georgette, dupioni)
  • Wool (merino, lambswool, cashmere, alpaca, mohair)

Cellulosics + synthetics

  • Viscose, modal, micromodal, Tencel/Lyocell, cupro, bamboo viscose
  • Polyester (PET, recycled rPET, microfiber)
  • Nylon (6, 66, recycled Econyl), acrylic, elastane/spandex

Knit constructions

  • Single jersey, interlock, rib (1×1, 2×2, 4×2)
  • French terry, fleece, pique, jacquard knit
  • Ponte, scuba, double-knit, bonded
  • Cable, fisherman, fully-fashioned, seamless

Woven constructions

  • Plain: poplin, voile, lawn, organza, taffeta, chambray
  • Twill: denim, gabardine, chino, drill, herringbone
  • Satin/sateen, dobby, jacquard, velvet, corduroy
  • Specialty: waffle, seersucker, oxford, tweed, melton
Q3
Quantity — how many for this run?
→ supplier.MOQ · supplier.MOQ ≤ buyer.qty
▾

Different garments have different MOQ economics. The MOQ tier matrix by product category drives factory-class + flexibility filter.

Sample / capsule

  • Cotton T-shirt (printed): 30–100 / color
  • Hoodies / sweatshirts: 100–300
  • Sweaters / knitwear (machine): 100–200
  • Woven dresses / blouses: 50–200
  • Leather goods (bags): 50–100

Standard runs

  • Cotton T-shirt (cut-and-sew): 500–1,000
  • Polo shirts: 300–500
  • Denim / jeans: 500–1,000 / wash
  • Activewear (poly-spandex): 500–1,000
  • Hats / accessories: 300–500

Scale

  • T-shirt mass: 1,000+ / color
  • Hoodies: 1,500+
  • Wool coats: 1,000+
  • Activewear: 2,000+
  • Athletic shoes: 3,000+

Flexibility heuristic

  • Flexible: vertically integrated small mills (Tirupur), wholesale-cluster factories (Guangzhou), seasonal-slack windows
  • Rigid: mass-production factories (Vietnam Tier-1), silk mills, fully-fashioned knit specialists
Q4
Channel + Price — where you sell, what you charge
→ supplier.FOB_range · back-calc target FOB
▾

Margin stack drives realistic FOB target. Buyer states retail + channel; AI back-calculates the FOB band.

FOB target by channel

  • DTC (Shopify own site): 12–20% of retail · brand 70–80% gross
  • Etsy: 25–40% · net 30–50% after fees
  • Amazon FBA: 18–30% · 25–40% net
  • Wholesale to indie: 25–35% · 40–55% on wholesale
  • Department / specialty: 15–22% · 30–45% net
  • Through middleman: 20–28% · 65–75% if DTC
  • Marketplace (Amazon/Tmall): 20–30% · 25–40% net
  • Luxury / premium boutique: 8–15% retail · 60–75% gross

Rule-of-thumb stack

  • Wholesale = 2.0–2.5× FOB landed
  • MSRP = 2.2–2.5× wholesale
  • DTC vertical retail = 4× FOB landed (Everlane / Warby model)
  • Example: FOB $10 → wholesale $25 → MSRP $60–75 traditional
Q5
Geography — where it's made + where it sells
→ supplier.country + supplier.top_markets
▾

Production-region preference filters factories by country. Sales-market answer auto-derives baseline compliance.

China — Pearl River Delta

  • Guangzhou: women's dresses, fast-fashion, wholesale
  • Shenzhen: premium / mid-premium women's wear
  • Dongguan: broad capability, jeans, outerwear
  • Foshan: denim, knitwear, underwear
  • Dalang: sweater capital — "1 in 6 sweaters globally"

China — Yangtze River Delta

  • Hangzhou: silk, premium women's, designer-caliber
  • Suzhou: silk heritage, embroidery, luxury silk
  • Shaoxing: Textile City of China — world's largest fabric market
  • Yiwu: >50% of China's leggings, ~70% seamless underwear
  • Huzhou: world's largest children's wear cluster

China — Fujian + Shandong

  • Quanzhou + Jinjiang: performance fabrics, athleisure, sneakers
  • Qingdao + Jinan: woolen fabric, suits, formal wear

India · Vietnam · Pakistan

  • Tirupur (IN): knitwear / cotton T-shirt capital, vertically integrated
  • Surat (IN): synthetics, polyester, sarees
  • Ludhiana (IN): woolens, sweaters, thermals
  • Ho Chi Minh region (VN): >60% of Vietnam exports — casualwear, US-bound
  • Hanoi region (VN): knitwear, outerwear, EU-bound
  • Karachi + Faisalabad (PK): cotton tee, baby clothing, knitwear
Q6
Timeline — when do you need it?
→ supplier.lead_time · supplier ≤ buyer.deadline − buffer
▾

Lead time filter + seasonality blackouts. CNY / Tet / Diwali drop regional capacity 6–8 weeks.

Lead time tiers

  • Sample lead: 7–21 days typical
  • Bulk lead: 45–90 days FOB · 60 days median
  • Express / Ready-to-Ship: in-stock SKU + minor modification
  • Hard launch deadline: framework backs out 2–3 week buffer for shipping + customs

Seasonality blackouts

  • Chinese New Year: 14–21 days closed · 6–8 week total disruption (mid-Jan–early Mar)
  • Tet (Vietnam): end-Jan / early-Feb · same pattern as CNY
  • Diwali (India): late-Oct / Nov · 2–3 weeks disruption
Q7
Compliance — required certifications
→ supplier.compliance_stack · supplier.certs ⊇ required_certs
▾

Buyer's market auto-derives required certs. 3-state filter (verified vs PDF-uploaded vs none). 15 major certs the framework recognizes.

Labor / social

  • BSCI: EU mass retail (H&M, C&A, Aldi)
  • SEDEX SMETA: Tesco, M&S, Walmart-adjacent
  • WRAP: Walmart, Target, Kohl's
  • SA8000: premium brands · most rigorous

Chemistry / environment

  • OEKO-TEX 100: EU + Japan + premium DTC
  • OEKO-TEX MIG: sustainable-positioned brands
  • GOTS: organic claim brands · baby / kids
  • GRS: recycled content claims
  • Bluesign: outdoor / performance
  • Higg FEM: Cascale members

Market access

  • CPSIA: mandatory for US children's wear
  • Prop 65: anything sold into CA
  • REACH: EU chemicals
  • ISO 9001 / 14001: process signals

Buyer-market → required certs

  • Mass US: WRAP + CPSIA + Prop 65
  • EU eco: GOTS or OEKO-TEX MIG + REACH
  • Outdoor / performance: Bluesign + Higg FEM
  • Premium DTC: OEKO-TEX 100 + (BSCI or SEDEX)
  • Children's: CPSIA + OEKO-TEX 100 Class I
Q8
Customization — logos, colors, sizing, prints, embellishment
→ supplier.vertical_integration · capability tier match
▾

9 customization vectors. Each drives sample iteration count, MOQ uplift, and which factories can fulfill.

Logo / branding

  • Method: woven label, printed (heat-transfer / TPU / silicone), embroidered (flat / 3D / puff), heat-transfer print, silicone patch, leather patch, debossed, sublimation, screen, DTG, foil, flock
  • Placement: chest, back of neck, sleeve, cuff, hem, side seam, pocket, hangtag

Color + sizing

  • Pantone TPX (paper), TCX (cotton), C / U, CMYK, custom lab dip
  • Lab dip: 2–3 iterations free, $30–80 per extra
  • Tolerance: ΔE <1 (premium), <2 (standard), <3 (mass)
  • Standard runs: XS–XL, XS–XXL, XS–XXXL · Petite, Tall, Plus, Maternity
  • Asian fit vs Western fit

Print + embellishment

  • Coverage: placement, all-over, repeat, engineered
  • Color count: 1, 2, 3, 4+ (drives cost)
  • Embellishment: beading, sequins, embroidery (flat / 3D puff / chenille), appliqué, laser-cut, smocking, pleating

Trims · wash · packaging

  • Buttons: corozo, horn, metal, polyester, shell, snap, hidden
  • Zippers: YKK (premium), SBS, KCC; exposed, invisible, two-way, water-resistant
  • Wash: garment dye, stone wash, enzyme wash, acid wash, distressed, vintage, brushed
  • Packaging: polybag, hangtag, tissue, branded inserts
Q9
Sample expectations — cost, lead time, iterations
→ supplier.sample_policy · cost + lead + iterations within tolerance
▾

#1 source of relationship breakdown in the first 30 days. Buyer expectations on sample cost, lead, and revisions filter against supplier's published sample policy.

Sample types

  • Counter sample: factory's reference, 1 piece, paid
  • Fit sample: for fit approval, made in available fabric
  • Pre-production sample (PPS): in actual fabric, final approval
  • Top of production (TOP): first off bulk, QA check

Cost + lead expectations

  • Counter sample: $20–80 (paid, refundable on bulk PO) · 7–14 days
  • Fit / PPS: $80–300 · 14–21 days
  • Custom-pattern development: $200–800 · 21–35 days · multiple rounds
  • Iterations: 2 included standard, 3 premium · +$ per extra
Q10
Trust + IP — comfort with sharing design files, language fit
→ supplier.lang + response_window + trader-flag
▾

Final filter — operational trust signals. Vertical integration tier, language fit, response-window match, and trader-vs-manufacturer distinction.

Vertical integration tier

  • CMT: you supply fabric; cheapest labor
  • Full-package: factory sources fabric; higher unit cost
  • Vertical-with-mill: factory has own mill; tightest control, small-batch friendly
  • Trading company: middleman, NOT manufacturer — auto-flagged

Language + response

  • Mandarin only → translation layer fires
  • Mandarin + English (export-side) → standard match
  • Cantonese (HK / Guangdong) → premium for HK-buyer fit
  • <1h: Fast Response tier · <12h: standard · >24h: red flag

Output preview

Brief: "200 oversized cotton tees, 240+ GSM, ivory + sage, $4 landed, US-bound, 8-week timeline." Three ranked matches with trade-off notes.

★ Best fit
Tirupur Knit Co. (India)
FOB: $3.20
MOQ: 200
Lead: 52 days
Vertical: Mill+factory
Best price match. Vertical mill + factory drops 25% off comparable Chinese FOB. Cotton-only specialty.
OEKO-TEX 100SEDEX
⚖ Quality match
Dongguan Apparel (China)
FOB: $3.80
MOQ: 300
Lead: 45 days
Vertical: Full-package
+18% FOB, but 3 sample iterations included and OEKO-TEX MIG verified. MOQ 100 above target.
OEKO-TEX MIGBSCI
⚡ Speed match
HCMC Garment Ltd. (Vietnam)
FOB: $3.50
MOQ: 250
Lead: 38 days
Vertical: CMT
Fastest turn — 38-day bulk lead. CMT means you supply fabric; saves 15% on FOB. US-export-experienced.
WRAP GoldCPSIA cleared
How everything you saw was built

Where the data came from + how the matches are picked

Nothing in the live demo is fabricated. Every supplier, FOB, MOQ, and audit signal traces back to a real listing on Alibaba's showroom pages. This page documents exactly how — so you can verify against source.

1 · Where the supplier graph comes from

📦
Source: Alibaba.com showroom + IndiaMART directory
→ ~19K records · ~12.5K unique supplier companies · 39 countries
▾

All supplier data was pulled from Alibaba's category showroom pages — the same pages a buyer would land on if they searched a category like "cotton-t-shirt" directly on Alibaba. We didn't build relationships, we didn't pay for premium data — we scraped what's publicly visible to anyone with a browser.

Per-record fields

  • Supplier name + Alibaba storefront URL (so you can click through to verify)
  • Country of origin
  • Years on Alibaba (supplier_year)
  • Trading-company flag (manufacturer vs middleman, auto-flagged in red)
  • Gold Supplier + Verified Supplier flags from Alibaba
  • Third-party auditor (TÜV Rheinland · SGS · Intertek · BV)
  • Promotion tags (Ready to Ship, etc.)
  • Reviews: star score + review count
  • Price ladder: tiered FOB by quantity bracket — verbatim from the listing
  • Provide products: the supplier's listed product strengths

What we did NOT scrape

  • Buyer messaging history (private)
  • Internal Alibaba metrics like "response time" beyond what's public
  • Anything paywalled or behind login
  • Image assets — text data only

Tools used

  • Scrapling (anti-bot Python scraping library)
  • Stealth browser fingerprinting + slow-paced pagination
  • Per-category resume logic (script can pause + restart without dups)

Run timestamp

  • Initial scrape: 2026-05-06
  • Latest data refresh: 2026-05-08
  • Coverage: 486 distinct category queries (cotton-t-shirt, hoodie, denim-jeans, etc.)
  • Pages per query: capped at the first 5 showroom pages (~150 listings/category) for breadth-over-depth

2 · How the chat extracts the brief

🤖
Multi-turn agent — Kimi K2 (text) + Claude Sonnet 4.5 (vision)
→ structured tool-call into 16+ dimensions, no free-text parsing
▾

The chat isn't a single "pull info from one big prompt" — it's a turn-by-turn agent walking the brand-brief framework one question at a time. The model decides each turn whether to ask the next question, render a fabric picker, or call match_brief with all dimensions filled.

Models in use

  • Text turns: moonshotai/kimi-k2-0905 via OpenRouter — picked specifically for native-fluent Mandarin sourcing-trade idiom (我司, 起订量, 大货, 打样, 克重)
  • Vision turns: anthropic/claude-sonnet-4.5 via OpenRouter — used when user attaches a sketch/mood image; reads silhouette, fabric texture cues, palette, brand tier

Interview order

  • 1. Product type
  • 2. Direct customer (gender, age, persona)
  • 3. Brand reference (Everlane, COS, etc.)
  • 4. Ideal retail price → FOB band auto-calculated (retail/4–retail/5)
  • 5. Audience deeper (lifestyle, climate, values)
  • 6. Fabric / texture (visual picker — see below)
  • 7. Color palette
  • 8. Visual references (optional, image attach)

Production decomposition

  • Complex briefs (e.g. "tees with college logos, no inventory") trigger a decomposition turn BEFORE intake continues
  • Project split into: factory_in_network · factory_out_of_network · talent_needed · external_service · user_handles
  • Honest framing — "We don't have POD fulfillment, here's Printful/Gelato/Printify" beats fake matches

Fabric picker (visual translation)

  • Non-fashion users can't articulate "240gsm fleece-back" — so we don't ask
  • Picker shows 3 layers per chip: plain English ("Buttery soft"), brand reference ("like Lululemon Align"), factory spec underneath ("260gsm nylon-spandex 4-way stretch")
  • User taps → factory spec flows into the brief as their answer
  • Curated for 8 highest-volume categories; generic lightweight/midweight/heavyweight fallback for the rest

3 · How the 3 matches are picked

⚖️
Hard filters → fit score → rank → top 3
→ deterministic, no LLM in the ranking loop
▾

After the brief is extracted, the supplier graph is filtered + scored client-side using a deterministic ranker — so the same brief always returns the same matches. No LLM is in the picking loop. This is a feature, not a limitation.

Hard filters (suppliers excluded if any fail)

  • Category match: supplier.search_query === brief.category
  • MOQ-fit: supplier.moq ≤ max(brief.qty × 1.5, 2000)
  • Has FOB data: supplier.fob_low_usd != null
  • Trading-company excluded: !supplier.is_trading_company (manufacturers only)

Fit score (higher = better)

score =
  trust + audit + years
  − fobGap × 1.5

where:
  fobGap = |fob − target| / target
  trust  = review_score × log10(reviews + 1)
  audit  = 0.4 if auditor exists, else 0
  years  = min(supplier_year / 10, 1.5) × 0.2

Why those weights

  • FOB gap is the single biggest factor (1.5×) — the buyer told us their target; missing it badly is a deal-breaker
  • Trust = review score × log(review count) — log dampens "5★ from 2 reviews" vs "4.7★ from 200 reviews"
  • Audit (+0.4) is a nudge, not a gate — many great factories don't have third-party audits, especially smaller mills
  • Years is mild (max 0.3) — capped because new factories aren't penalized as harshly as fly-by-night

Top 3 picks

  • Sorted by score, top 3 returned
  • Each card carries a trade-off note ("+18% FOB but 3 sample iterations included" / "best price but trading-company") — the AI explains the trade so the buyer doesn't have to read tea leaves
  • Position 1 = ★ Best fit · Position 2 = ⚖ Quality match · Position 3 = 💡 Alternative

4 · The first-message draft

✉️
Brand-led intro · EN + Mandarin sourcing-trade idiom · auto-translates supplier replies
→ generated once per brief, ready to copy + send
▾

Once a top match is picked, the agent drafts the first outreach message in English + Simplified Chinese. The pattern follows what 30-year Asia sourcing operators actually use: open with brand context (who/why/what need), THEN the spec, THEN 4 numbered questions, close warm. Generic supplier reference so the message can fire to any of the top 3.

Why this matters

Suppliers triage hundreds of inquiries per week. A brief that opens with "we're sourcing 200 cotton tees" looks like a tire-kicker. A brief that opens with "we're building a Singapore-based basics label for sustainability-conscious professionals…" reads as a real brand who's done the work. Higher-quality factories ALWAYS prioritize real brands.

5 · What's NOT in this demo (yet)

Honesty about the gaps — what's still v0.5 vs v1.0 production:

Coming next

  • Sentiment vetting from Xiaohongshu / 小红书 (Chinese buyer reviews)
  • TikTok Shop signal for D2C-fluent suppliers
  • Real-time response-window data (currently only static metadata)
  • WhatsApp + WeChat handoff after first message
  • Multi-supplier match for decomposed projects (today: top 3 in primary category only)
  • Talent layer (pattern designers, branding, packaging) — currently flagged but not matched

Already live in this demo

  • Alibaba 14,030 listings + IndiaMART 4,917 listings · ~12.5K unique companies · 39 countries
  • Multi-turn guided chat (8-question interview)
  • Visual fabric picker (8 curated category libraries)
  • Production decomposition (project supply chain breakdown)
  • EN + ZH first-message generation (sourcing-trade idiom)
  • Image-attach support (vision-model fabric reading)
  • Out-of-network honesty (suggests adjacent categories or talent partners)
Live data · Alibaba + IndiaMART · cleaned per Frenzee framework

— real supplier listings across two platforms

Pulled live from Alibaba.com showroom pages AND IndiaMART directory pages (the dominant B2B sourcing platform in India) using Scrapling. Cleaned via the Frenzee framework — 10 buyer-intent dimensions mapped to supplier-schema fields. factories.json · cleaned combined · raw Alibaba · raw IndiaMART

—
Total listings
—
Unique supplier companies
—
Categories covered
—
With third-party trust signal
—
Alibaba listings
—
IndiaMART listings

Country breakdown — where suppliers are based

Top sourcing geographies in the dataset
loading…
Supplier ↕ Category ↕ Country ↕ Years ↕ FOB ↕ MOQ ↕ Auditor ↕ Tags Score ↕
Loading supplier graph…