THINVENT TECHNOLOGIES · WWW.THINVENT.IN · SYSTEM REFERENCE
The website, drawn as a board
We build fanless computers, so we have drawn our website like one of our own boards:
each subsystem is a module and each data flow is a trace. The teal signal follows one visitor from ad
click to delivered order, and back to the ad platform's bidding system.
data flow the closed loop detailed below
TEAL PULSES = ONE VISITOR: AD CLICK → PAGE → CONTACT → TICKET → CONVERSION REPORTED → AD BIDDING UPDATED
245,638
SEO pages live
224
parts in productdb
20
languages
15
product families
12
customer case studies
U1
Product core: the SKU is a sentence
One database (224 parts, 10 categories) generates every page, price, datasheet and feed.
A product is a compatible combination of parts, written as a code that every part of the system parses:
EACH FIELD IS A REAL PART · EACH PART LISTS WHAT IT ALLOWS, PROVIDES AND CONTAINS · ONE VALIDATOR CHECKS PAGES, CHAT AND CHECKOUT
PRICE PATH — computed live, never stored
Part costslive from the ERP
→
Σ SKU costsummed per configuration
→
Pricing modelper product class
→
₹ web pricein the visitor's currency
ONE DATABASE, SIX SHELVES
/p/<sku>product page
/f/<family>configurator — adapter auto-picked by power budget
/gs/<sku>priced page for ads
/ds/<sku>.pdfdatasheet on demand
Merchant feed
Amazon listings
U2
The SEO factory
The 245,638 pages are not written by hand. A pipeline reads search queries, groups them by
meaning, matches them to products and builds a page per group, and then reloads the live index.
QUERY SOURCES
Search Consolewhat we rank forAds search termswhat we pay forKeyword ideaswhat we could rank forLive site searcheswhat visitors type here
THE CONVEYOR
1
Combine
merge and dedupe all sources
2
Embed
every query becomes a meaning-vector
3
Understand
"16gb fanless pc" → RAM=16 + fanless
4
Cluster
similar queries share one page
5
Match
clusters → the SKUs that answer them
6
Build
245,638 query pages + sitemaps → Google
7
Link + write
related-search links · AI descriptions
8
Go live
vectors loaded into the search index
RUNS ON RENTED CLOUD COMPUTE · INCREMENTAL · LOW-TRAFFIC PAGES PRUNED BY ACTUAL VISIT COUNTS
U3
Live search: milliseconds, not databases
Visitor typessearch box or /q/ page
→
Understandwords → filters (RAM, CPU, fanless…)
→Vector indexmeaning-based search in memory→
Resultsproducts + live prices + facets + related searches + autocomplete
A dedicated search service responds to every keystroke without loading
the main site, and every search it receives becomes input for the next pipeline run.
U4
The ads robot: campaigns tuned by API
Google Ads APIprogrammatic account access⇄
COLLECT · nightlyperformance + search terms → local evidence base
Built to orderparts → assembly → thermal test → OS → QC → packing
→Ship & tracklive courier status on your order page
TO REQUEST A QUOTE, SKIP PAYMENT · A SALESPERSON HANDLES YOUR REQUEST IN EITHER CASE
Fulfilment — from our own order history
Stage
Typical
Notes
Order confirmed → dispatched
about a week
larger or custom builds take longer — sales confirms at order time
Courier, metro cities
2–4 days
tracked door delivery
Courier, rest of India
up to a week
tracked door delivery
THE CHATBOT QUOTES THESE SAME NUMBERS, WHICH ARE MEASURED FROM ORDER HISTORY
U9
The mailer: campaigns that clean up after themselves
Subscriber intakecustomers, enquiries, exhibitions — verified before use
→Campaignsthrottled sending→
Opens & clicksmeasured from our own server logs
→Suppressionbounces, complaints and unsubscribes removed automatically
Re-engagement emails go only to inactive leads, and never to anyone
who opted out, complained or bounced.
U10
Content, languages & case studies
20 languageshybrid translation systemDeepSeekdrafts page copyGeoIPcountry → currency10 currenciesdaily rates, billing in INREmbeddingsmeaning as vectors, everywhere
CASE-STUDY LINE
Real deploymentsactual customers and their installations
→AI draftsfrom first-party facts only→12 case studiespublished under /a/→
× 20 languagesthrough the same translation line
U11
Who runs what
Public web servernginx → Flask site · search service · livechat — behind CloudFront
ERP serverorders, manufacturing, finance, CRM back office, scheduled jobs
Rented computerented by the hour for the SEO pipeline, then shut down
StoresDynamoDB · Redis · S3 · shared storage
CloudFrontnginxFlaskDynamoDBRedisS3
J1–J22
The connector panel
All services connected to the website. Teal: visitors and data in · blue: operations
and payments · amber: AI.
Google Ads API
collection · negatives · ad groups · conversion upload
Search Console
query data in · sitemaps out
Merchant Center
shopping feeds from productdb
Amazon SP-API
marketplace listings, valid configs only
LinkedIn API
lead conversions reported back
Razorpay
payments + instant confirmation
Courier APIs
live shipment status on your order page
Exotel
phone-call enquiries into CRM
Google Chat
notifies the team of every lead
Email ingest
mailbox enquiries become tickets
Transactional mail
quotes, order updates, auto-replies
WhatsApp
one-tap contact, journey-tracked
Anthropic Claude
livechat engine + the agents that maintain this system
Amazon Bedrock
fallback chat engine · AI search assist
DeepSeek
drafts page copy at scale
Sentence-transformers
the embedding model behind search + SEO
AWS Textract
visiting-card OCR at exhibitions
AWS Translate
part of the 20-language system
MaxMind GeoLite2
country-level location → currency
Exchange rates
daily, 10 display currencies
CloudFront + S3
CDN, media, cache-busted assets
DynamoDB · Redis
system of record + in-memory index and cache
GENERATED FROM THE CODEBASE · NUMBERS ARE MEASURED FROM LIVE DATA
Logos are simplified marks of their respective owners, used to identify the
integrated services. More depth on each subsystem: see the technical documentation index.