Retail Logistics: Real-Time Data Platform Processing 2M+ Events/Day

Client: Retail logistics platform · 800+ stores across India · ~$180M GMV

Engagement: 12-week lakehouse build · November 2025–February 2026

The Challenge

Fragmented data infrastructure: 14 source systems, 12-hour data latency, no unified analytics layer. Operations teams making decisions on stale data. The 2025 holiday season exposed the cost — $2M+ in inventory mismatches.

Approach

Sample: Inventory Reconciliation dbt Model

-- models/marts/inventory_realtime.sql
{{ config(
    materialized='incremental',
    unique_key=['store_id', 'sku'],
    on_schema_change='sync_all_columns',
    incremental_strategy='merge'
) }}

with pos_movements as (
    select
        store_id,
        sku,
        sum(quantity_change) as net_movement,
        max(event_ts) as last_event
    from {{ ref('silver_pos_events') }}
    {% if is_incremental() %}
        where event_ts > (select max(last_event) from {{ this }})
    {% endif %}
    group by 1, 2
),

current_inventory as (
    select
        store_id,
        sku,
        on_hand_qty,
        reserved_qty,
        on_hand_qty - reserved_qty as available_qty,
        last_count_ts
    from {{ ref('silver_inventory_snapshots') }}
)

select
    coalesce(p.store_id, c.store_id) as store_id,
    coalesce(p.sku, c.sku)            as sku,
    c.available_qty,
    p.net_movement,
    c.available_qty + coalesce(p.net_movement, 0) as projected_qty,
    case
        when c.available_qty + coalesce(p.net_movement, 0) < 0 then 'OVERSOLD'
        when c.available_qty + coalesce(p.net_movement, 0) < 5 then 'LOW'
        else 'OK'
    end as status,
    greatest(p.last_event, c.last_count_ts) as last_updated
from pos_movements p
full outer join current_inventory c
    on p.store_id = c.store_id and p.sku = c.sku

Results

“We went from 12-hour stale data to real-time operational dashboards in under 12 weeks. The platform Ohveda built is the foundation of every decision we make today.” — Rahul Sharma, Head of Engineering