$ whoami — available for consulting

Hello, World. I'm Abhishek
Kushwaha

Roads in. Routes out. I'm a software engineer who'd rather engineer than just build — I design, scale and productionize self-hosted routing engines (Valhalla & OSRM) so your team owns its routing stack instead of renting it from Google.

⚑ Work with me </> What I do

10+ yrs engineering software
5+ yrs building logistics @ FarEye
$0 Maps API bills after migration

abhishek@routing-engine: ~

$ neofetch
  o---o      abhishek@routing-engine
 /     \     ------------------------
o   o   o    Role:    Principal Engineer @ FarEye
 \   |  /    Domain:  Routing · Optimization · Geo
  o--o--o    Engines: Valhalla, OSRM, GraphHopper
     |       Stack:   Ruby on Rails, Python, Go, C++, K8s
     x       Data:    OpenStreetMap · PBF · GTFS
              Scale:   last-mile telematics @ fleet scale
              Uptime:  shipping software since 2015
$ ./optimize --routes --realtime █

Valhalla·OSRM· OpenStreetMap·GraphHopper· C++·Python· Go·Docker· Kubernetes·PostGIS· GTFS·Isochrones· Map Matching·Matrix API· Routing Engine·GPS Pings·Slot Booking·Helm·HPA·Ruby on Rails·Ocean Tracking·Event-Driven·VRP·

// what I do

Routing infrastructure, productionized.

Anyone can docker run a routing container. Making it survive real traffic, real map updates and real customers is a different sport. That's the part I've done over and over — and the part I'll do for you.

⚑

Self-host Valhalla & OSRM

Stand up your own routing engine on OpenStreetMap data — route, matrix, isochrone, map-matching — running in your cloud, on your terms.

kill the Google Maps bill

⇄

Distance & time matrices at scale

Sub-second N×N matrices feeding VRP and dispatch. I've tuned Valhalla and OSRM for high-throughput matrix workloads without the latency cliff.

high-throughput

⚙

Custom costing & truck routing

Vehicle restrictions, weight/height limits, turn penalties, time-based access, hazmat and last-mile constraints baked into the graph.

domain-aware

↻

OSM data & tile pipelines

Automated PBF extracts, tile/graph builds and scheduled refreshes so your map is never months stale — with zero-downtime swaps.

always fresh

☍

Scale, HA & observability

Kubernetes autoscaling, health checks, caching, load tests and dashboards — the boring reliability that keeps routing up at 3am.

production-grade

➝

Migrate off Google / Mapbox

A measured migration path from paid APIs to self-hosted routing — parity tests, cost model, cutover plan, no surprises.

own your stack

📡

GPS ping ingestion at scale

Streaming device pings into real-time vehicle state — moving, idle, stopped — with clean storage, geofencing and event triggers that don't fall over under load.

real-time telematics

📅

Delivery slot-booking engines

Capacity- and feasibility-aware time-slot booking wired straight into routing — so the promise you make at checkout is one the fleet can actually keep.

capacity aware

☢

Kubernetes for routing workloads

HPA autoscaling, Helm-packaged deploys, resource tuning and zero-downtime rollouts — routing infrastructure that scales up for peak and back down for the bill.

HPA · Helm

// selected work

Case studies from the routing trenches.

A few representative problems I've solved with self-hosted routing — anonymized, but the engineering is real.

Valhalla

Cutting the Google Directions bill to near-zero

Problem A last-mile fleet was metered on every Google Directions call — costs scaled linearly with growth, and truck constraints were impossible.

What I did Stood up Valhalla on OpenStreetMap in their cloud, built a custom truck costing profile (weight/height/turn rules) and an automated tile-refresh pipeline, all containerized on Kubernetes.

~90%lower routing spend, self-owned stack

OSRM

Sub-second matrices feeding the optimizer

Problem Route optimization was starved — the distance/time matrix API was too slow and too costly to compute the N×N inputs a VRP solver needs.

What I did Deployed OSRM with the MLD pipeline, tuned it for high-throughput /table queries, and put it behind autoscaling with caching so peak dispatch never stalls.

<1slarge matrices, high-RPS, no latency cliff

Routing × Slot booking

Promising delivery windows the fleet can keep

Problem Customers picked delivery slots the operation couldn't actually serve — broken promises, failed deliveries, angry support tickets.

What I did Built a capacity- and feasibility-aware slot-booking engine that checks each slot against live routing feasibility before it's offered — so the promise at checkout is one the route can honor.

fewerfailed promises, tighter delivery windows

// about me

I've been living in the routing graph for years.

At FarEye I architect the real-time engine behind last-mile logistics — GPS-ping processing, alerts, slot-booking and multi-point routing for real fleets on real roads. Alongside that I take Valhalla and OSRM from raw OSM all the way to hardened, autoscaling production on Kubernetes.

Outside my day job I take that same playbook to other teams as a consultant — helping startups and enterprises skip the year of trial and error and get a routing stack they actually control.

Read the full story →

10+years shipping software

5+years at FarEye

2015shipping since

2engines mastered

// currently

What I'm on right now.

Updated Aug 2026 · Noida, India

⚙ Re-platforming FarEye's real-time logistics product into a faster, cleaner stack.
⚑ Taking on select Valhalla / OSRM consulting engagements on the side.
⇄ Tinkering with faster distance-matrix pipelines and custom costing models.
☍ Going deeper on Kubernetes autoscaling for spiky, real-time routing workloads.