Hatim Patwa
Hire me

India · IST

Hatim PatwaSDE-3 · Forward Deployed EngineerI turn ambiguous problems into production systems.

I drop into unfamiliar production systems, find what's actually broken, and ship the integration, data pipeline or AI infrastructure that fixes it.

See the workGet in touchDownload resume

The 10-second version

years shipping production software, since 2021
5+
industries: procurement, legal tech, mining, real estate, blockchain
5
Forward Deployed Engineer at Systango
SDE‑3
Certified Architect: Foundations (Anthropic)
Claude
Core stack

TypeScript · Node.js · NestJS · Python · Rust · PostgreSQL · AWS

Selected work

Client engagements, anonymized. Outcome first.

  1. Legal tech · US law firm

    Found why legal data arrived 19 days late.

    Dagster pipelines, a Python monorepo and TypeScript ETL pulling Zoho and QuickBooks into Postgres, MongoDB and Snowflake.

    median Zoho → Postgres replication lag, diagnosed
    19d
    of legal matters unfindable, traced to one defect
    13.8%
    • Mapped how data really flowed across a multi-repo system.
    • Built reconciliation tooling that improved data quality and observability.
    • PostgreSQL
    • TypeScript
    • Python
    • Zoho APIs
    • Snowflake

  2. Legal tech · Database migration

    Moved reporting off Azure SQL onto Postgres on AWS.

    Reporting databases on Azure SQL Server, moving to PostgreSQL on AWS RDS.

    • Built the migration tooling end to end.
    • pg_cron jobs with timezone-correct scheduling, IST → UTC.
    • Tunneled access to RDS.
    • pgloader
    • Python
    • PostgreSQL
    • AWS RDS
    • Docker

  3. AI infrastructure

    LLM agents on production data. Read-only, by design.

    Agents needed production reporting data without anyone having to trust a prompt.

    • Rust MCP servers with secure, read-only data access.
    • Python orchestration and an internal AI analyst interface.
    • Least privilege and enforced boundaries, not prompt-level trust.
    • Rust
    • Model Context Protocol
    • Python

  4. Procurement SaaS · Data export

    Exports of any size, in constant memory.

    An enterprise procurement platform needed every dashboard downloadable as Excel, without blocking the API or loading whole datasets.

    report types on one streaming pipeline, each a single strategy class
    8
    files on disk: rows stream from Postgres straight into Blob Storage
    0
    • Batched reads into an object-mode stream with backpressure, through ExcelJS's streaming writer to Azure Blob.
    • Strategy and Factory: one export type maps to several strategies, so one request builds a multi-sheet workbook.
    • NestJS
    • Node streams
    • ExcelJS
    • PostgreSQL
    • Azure Blob

  5. Procurement SaaS · Enterprise SSO

    One sign-in page for every enterprise identity provider.

    Corporate users had to land on their own company's identity provider; everyone else kept email and password, in the same journey.

    orchestration steps in one custom Azure AD B2C journey
    15
    code changes to onboard a new enterprise: the domain mapping is config
    0
    • Home realm discovery: the email domain routes users to their company's IdP, with login_hint passed through.
    • Just-in-time provisioning: B2C calls NestJS APIs to check, map and create the user on first login.
    • Email OTP for MFA, and password-expiry enforcement through Microsoft Graph.
    • Azure AD B2C
    • IEF custom policies
    • OIDC
    • NestJS
    • Microsoft Graph

Platforms I helped ship.

Named products, built from the inside over months, not weeks.

  • Nomia homepage
    Homepage
    Enterprise procurement · 2025–26

    Nomia

    Autonomous procurement for enterprises.

    RFQs, supplier onboarding, orders and invoicing, with separate portals for customers, suppliers and admins. NestJS microservices on Azure, in an Nx monorepo.

    • Streaming export engine: Postgres → multi-sheet Excel → Blob Storage, in constant memory. 8 report types.
    • Azure AD B2C SSO: domain-based routing to corporate IdPs, just-in-time provisioning, config not code.
    • NestJS
    • Nx
    • Node streams
    • PostgreSQL
    • Azure AD B2C
    • Dynamics 365 BC
  • MineTrak homepage
    Homepage
    Mining equipment SaaS

    MineTrak

    Offline-capable inspections and compliance.

    Operators scan an equipment QR code and inspect it on configurable forms; defects become severity-graded tickets, with MSHA and OSHA checklists built in.

    • JSON-schema inspection forms: photos and signatures, offline sync from the field, audit-ready exports.
    • Org → site tenancy via Prisma middleware; one codebase serves the API and an email/SMS worker on Pub/Sub.
    • NestJS
    • Prisma
    • PostgreSQL
    • GCP Pub/Sub
    • Docker
    • Multi-tenant
  • Legal tech
    Immigration legal tech

    4iiz

    Software behind a US immigration law firm.

    Case management, consultant routing, a CRM-to-Postgres data platform and AI call coaching for a high-volume practice. Django, Dagster and Angular on AWS.

    • Consultant routing: empirical-Bayes hire rates minus live workload, with a three-level fallback cascade.
    • Zoho → Postgres sync: exactly-once keyset paging, schema-drift DDL, flat memory (272 → 90 MB).
    • Python
    • Django
    • Dagster
    • Angular
    • PostgreSQL
    • AWS

Toolkit

TypeScript and Node.js daily. Python and Rust when the problem calls for it.

Core
  • TypeScript
  • JavaScript
  • Node.js
  • NestJS
  • SQL
Also strong
  • Python (Django, FastAPI)
  • Angular
  • React / Next.js
  • Rust
Data
  • PostgreSQL
  • MongoDB
  • MySQL
  • Redis
  • Supabase
  • Azure SQL Server
  • ETL/ELT
  • Data reconciliation
Cloud and delivery
  • AWS
  • Azure
  • GCP
  • Vercel
  • Docker
  • GitHub Actions
  • CI/CD
  • Playwright
  • Nx monorepos
AI
  • Anthropic Claude API
  • Model Context Protocol (MCP)
  • LLM agents
  • RAG
  • Claude Code
Practices
  • System design
  • Multi-tenant SaaS
  • Microservices
  • Root-cause analysis
  • Release coordination

How I work

Root causes over quick patches. Enforced guardrails over hoping a prompt behaves.

  1. 01

    Start from the real problem

    Map the domain and the data flow before writing code.

  2. 02

    Own it end to end

    Design, build, release coordination, production debugging.

  3. 03

    Make it observable

    Reconciliation, scheduled jobs and data-quality checks surface problems early.

  4. 04

    Constrain AI by design

    Agents get least-privilege, read-only access with enforced boundaries.

From PHP to production AI.

Started in 2021 building PHP and MySQL apps. Today I work inside client production systems across procurement, legal tech and mining.

  1. 2021

    Web Developer

    Saify Technologies

    PHP · MySQL · JavaScript

  2. 2022

    Software Developer

    CDN Software Solutions

    Angular · Node.js · SQL · Web3.js

  3. 2025

    SDE-3 · Forward Deployed Engineer

    Systango · now

    Integrations · Data · AI infrastructure

Credential

Claude Certified Architect: Foundations

Anthropic

Contact

Got a system nobody fully understands? Let's talk.

hatimpatwa5253@gmail.com