Skip to content
J—S

I build productionsystems that don'tbreak at scale —and put AI inside them.

Senior full-stack engineer · Tech lead · Systems architect
6+ years · Distributed systems · Real-time · Multi-tenant SaaS
Karachi → Singapore · Germany · UK · Remote

Fig. 00 — Signal at rest

Scroll

§ 01 — Proof

FIG. 01

Proof

Production systems shipped

10+

Engineers led across squads and platforms

SPLC codes resolved

214K+

Railroad location codes per import cycle

Manual work removed

~2h

Saved per person per day via LLM pipelines

Developers directed

20+

Across web, mobile, and SaaS products

§ 02 — Featured work

FIG. 02

Featured work

FIG. 02.1

CSVEXCEPTIONOD PAIRMAP

Everest Railcar Services

Freight rail SaaS · Dispatch · Real-time tracking

Problem

Rail operators processed daily CSV trace reports by hand. Backlogs stretched into days. Exception handling fell through the cracks. Fleet visibility lived in a spreadsheet.

Solution

I designed a 3-queue BullMQ pipeline — csv-import, exception-evaluation, od-recalculation — with idempotent job keys and atomic Prisma transactions.

Outcome

Daily manual processing went to zero. Each import cycle resolves 214k+ SPLC location codes.

Node.js · TypeScript · PostgreSQL · Redis · BullMQ · Prisma · AWS

Read full case study →Read full case study →

FIG. 02.2

ADMINCOACHSTUDENTRBAC

5 Peaks Youth Solutions

FERPA-compliant EdTech SIS

Problem

Administrators, coaches, and students needed one product. FERPA records could not cross roles or tenants.

Solution

I architected three portals on one NestJS domain. JWT roles, tenant isolation, and live presence over WebSockets.

Outcome

Admin, coach, and student surfaces share one RBAC model. Student records stay isolated.

Next.js · NestJS · PostgreSQL · Prisma · Redis · JWT

Read full case study →Read full case study →

FIG. 02.3

ROOMSOCKETSCORENOTIFY

ShackWise

Collaborative PropTech

Problem

Buyers, partners, and agents scored homes in separate threads. Ratings drifted. Search stayed manual.

Solution

Shared rooms sync over Socket.IO. BullMQ runs scoring and notification jobs off the request path.

Outcome

Concurrent ratings land in one room state, with a priority-weighted home score.

React 18 · Node.js · MongoDB · Redis · Socket.IO

Read full case study →Read full case study →

FIG. 02.4

INTAKERANKOFFERSHIFT

Azara Health Care

Care coordination & dispatch

Problem

Coordinators matched caregivers by phone and spreadsheet. Shift assignment lagged behind intake.

Solution

Intake, Distance Matrix ranking, and queue-backed offers sit on Next.js, Node, PostgreSQL, and Prisma.

Outcome

Ranked matches replace manual search. Four roles share one shift lifecycle.

Next.js · Node.js · PostgreSQL · Prisma · Distance Matrix

Read full case study →Read full case study →

FIG. 02.5

INPUTBUDGETJSONZOD

Production AI / LLM Tooling

Enterprise automation

Problem

Teams wanted model help inside existing tools. Malformed JSON and blocked request threads were the failure mode.

Solution

Sanitized input, a token budget, JSON-mode completion, a Zod guard with two repair retries, then an ACID commit.

Outcome

About two hours a day came back per person on the internal tool.

TypeScript · Zod · BullMQ · Redis · PostgreSQL

Read full case study →Read full case study →
View all projects →View all projects →

§ 03 — Systems

FIG. 03

Systems

Service topology for queuesAPIBULLMQREDISDLQ

§ 04 — AI in production

FIG. 04

Deterministic pipelines, not prompt roulette.

  1. Sanitized input

  2. Token budget

  3. LLM · JSON mode

  4. Zod guard

  5. ACID commit

Fallback sits under the Zod guard. Two retries, then the safe path.

Why deterministic

A model response is untrusted input. I parse it, reject it, or repair it before any write.

The request thread never waits on a long completion. BullMQ holds the job.

Temperature stays low. The schema is the contract.

Failure modes handled

  • Malformed JSON
  • Token overflow
  • Provider timeout
  • Schema drift
  • Duplicate job keys

Measured impact

~2 hours/day

Recovered per engineer on the internal tool.

§ 05 — Experience

FIG. 05

Experience

  1. Nov 2024 — Present

    Netpace Pvt Ltd · Senior Software Engineer & Tech Lead

    Karachi, PK

    I lead 10+ engineers across freight, PropTech, and healthcare. I own architecture review, sprint scope, and the LLM tools those squads actually run.

    Selected work

    • Everest Railcar
    • ShackWise
    • Azara Health Care

    Led 10+ engineers · architected 3 flagship platforms · shipped LLM automation saving ~2h/day

  2. Apr 2024 — Oct 2024

    Concept Recall · Senior Software Engineer

    Karachi, PK

    I directed delivery across web, mobile, and a multi-tenant SaaS. The core was Node, Next.js, PostgreSQL, Redis, and BullMQ.

    Selected work

    • Multi-tenant SaaS
    • Enterprise API layer

    Directed 20+ developers across web, mobile, and SaaS

  3. Nov 2022 — Apr 2024

    Sunbonn · Senior Software Engineer

    Remote · UAE clients

    I shipped MERN and React Native products for commercial clients, from architecture through release.

    Selected work

    • Hana Water
    • MJ Jafferjees

    Owned full-stack delivery for commercial mobile platforms

  4. Jan 2022 — Oct 2022

    Kahoo Digitals · Mid-Level Software Engineer

    Karachi, PK

    I built Node and Express APIs and React interfaces, including the MaxTrace telemetry app.

    Selected work

    • MaxTrace

    Shipped React Native telemetry on Node APIs

  5. 2020 — 2021

    Technado & 5 Cube · MERN Stack Developer

    Karachi, PK

    I wrote Node and Express APIs, tightened queries, and added Redis caching on early product work.

    Selected work

    • API throughput
    • Redis cache

    First production MERN systems · queries, cache, and test coverage

BSc Computer Science · Iqra University, Karachi

§ 06 — About

FIG. 06

About

I have spent six years putting TypeScript systems into production: freight, schools, property, and care. The work is queues, tenancy, auth, and the moment a model output has to become a database row. I lead squads of ten to twenty. I write the design doc, review the pull request, and stay on the incident. Karachi is home. I am ready to relocate to Singapore, Germany, or the UK.

Operating principles

  • Ship correctness first. Optimize second.
  • Every queue needs a dead-letter story.
  • If it can't be observed, it isn't in production.

Currently learning

  • Deterministic evals for LLM pipelines
  • Multi-region queue topology

Outside the terminal

Evenings go to long walks and the odd history book. The terminal closes.

§ 07 — Contact

Let's build something that holds.

Direct