Case file 03 · Head of Engineering

BlueRickshaw

The operating system for importing food & drink into India — from a supplier abroad to a kirana shelf.

Role
Head of Engineering
Period
Oct 2026 – present · built from May 2026
Cargo
~80 FastAPI routers · Gemini 2.5 intake · voice agent · MCP
Status
In production
CONSIGNMENTBR · IN
Aerial view of a container port with stacked shipping containers (placeholder image) Placeholder — swap for a real screenshot
FromBrands abroad
ViaBR USA → BR India
ToKirana & restaurants
A shipping-label styled card summarising the BlueRickshaw import chain.

The problem

A jar of imported jam changes hands five times before it reaches a shelf.

Every hand has its own spreadsheet, its own WhatsApp group, its own pile of PDFs. Orders arrive as emails. Stock expires in the back room. Nobody knows when the ship lands.

I'm building one system for all of it. Orders, catalogue, procurement, invoices, shipments and deliveries live in one place, and the AI parts do the boring reading so people don't have to.

How the goods move

One route. Five stops.

Scroll to ship it

  1. 01Brands abroadUS, UK, EU, Asia suppliers
  2. 02BlueRickshaw USAConsolidates & ships
  3. 03BlueRickshaw IndiaClears, stores, sells
  4. 04DistributorsRegional buyers
  5. 05Kirana & restaurantsThe shelf
Simulation

Email in. Draft PO out.

A simulated, in-browser version of the document intake. In production, Gemini 2.5 reads purchase orders from email, PDF or spreadsheet and turns them into priced draft POs. Here a small matcher plays its part. Edit the email and run it.

Inbox · 1 new order
Fromorders@example-distributor.in
SubjectPO for next week — please confirm
  1. 01Email received
  2. 02Gemini 2.5 reads it
  3. 03Match to catalogue
  4. 04Price the draft
  5. 05Human approves

Draft purchase order

PO · — — —

Example prices

Nothing here yet.

Run the intake to read the email.

Waiting for an email.

Around the intake

The rest of the back office, rebuilt.

VoicePriya

"Priya" takes orders over the phone, in Hinglish.

Shop owners just call. A voice agent built on LiveKit, Sarvam STT, Deepgram and Groq Llama 3.3 70B turns the call into an order.

Simulated call
    LiveKitSarvam STTDeepgramGroq · Llama 3.3 70B

    Sea and air shipments, tracked by webhook.

    Carriers push events; the shipment page updates itself. Example events below.

      A stockroom app that picks by expiry.

      Scan a barcode; the app sends you to the crate that expires first (FEFO). Example dates.

      6 crates on the shelf.

      Staff run operations from Claude.

      An MCP server exposes the app to Claude, acting as the signed-in person.

      When does the next sea shipment land?

      tool shipments_arriving()

      One container is due at the port this week. Want me to list the distributors waiting on it?

      ~80 FastAPI router modules

      Backend surface

      orderscatalogue & searchinventoryprocurementinvoicesshipmentsdeliveriesanalyticsWhatsAppvoiceMCP

      Three environments, watched.

      • prodDocker Compose + Nginx
      • sandboxDocker Compose + Nginx
      • devDocker Compose + Nginx
      PrometheusGrafanaLoki

      Numbers

      • ~80FastAPI router modules
      • 5links in the import chain
      • 3intake formats: email, PDF, spreadsheet
      • 3environments: prod, sandbox, dev

      Reel

      Ten seconds of cargo.

      Placeholder — swap for a real product recording

      Stack

      What it's packed in.

      • Python
      • FastAPI
      • PostgreSQL 16
      • Redis
      • Elasticsearch
      • MinIO
      • Next.js 15
      • Gemini 2.5
      • LiveKit
      • Docker
      • Grafana
      • pytest
      • Playwright

      Next project · 04

      CLUE AI

      A syllabus-aware AI tutor for IB students, grounded in their textbooks and their own past lessons.