Jamshaid Ali
Senior Software Engineer · Lahore · Open to remote

I build Python backends and AI agents that ask before they act.

6+ years shipping production systems with Django, FastAPI, PostgreSQL and AWS. Today I build MCP servers and AI agents that do real business work, with a person approving every change that matters.

fundraising-agent · investor CRMexample run
Team member
Add Helio Climate Fund to the CRM. Deep search.
Agent
  • Searched the CRM by name and website: no duplicate
  • Web search: firm profile and focus
  • Coresignal: team data (1 search, 2 profiles)
NameHelio Climate Fund · website
StageTriage
FocusSeed–Series A, energy transition · web
Team12 people, 3 partners · Coresignal
TicketDA VERIFICARE · no source
Waiting for your approval. Nothing is saved yet.
Tools used: web search + Coresignal (1 search, 2 profiles)
6+ yrsbuilding production backends
5 min → 30 sdata processing after query tuning
8+Kubernetes clusters upgraded
−40%operational cost from database consolidation
01 / Selected work

Things I've shipped

AI agents and MCP servers in production, plus the backend and infrastructure work behind them.

AI agent · ClickUp + MCPOptimaGeeks · Oct 2026 – now

AI fundraising agent on a live investor CRM

An agent inside a climate-tech company's ClickUp workspace that finds investors, keeps a CRM of about 170 VCs current, writes investor briefs and drafts outreach emails. The demo at the top of this page is how it works.

Live on the real CRM: new VCs arrive with sources, no duplicates, and every change approved by the team.

  • Wrote the agent's instructions: role, scope, hard rules (no deletes, no duplicates, no moving a stage backward) and hand-offs
  • Human-in-the-loop writes: a duplicate check, a preview with a source for every field, and a save only after a yes
  • Connected ClickUp, web search and the Coresignal MCP server, with paid lookups opt-in only and capped per investor
  • A "Tools used" line on every answer, so the team always knows where data came from
  • Shortlist and article-to-CRM flows, email-to-task intake, and English/Italian replies
AI agentsMCPCoresignalClickUpHuman-in-the-loop
Multi-agent · MCP · OAuthOptimaGeeks · Jun 2026 – now

AI marketing & business operations platform

Helps business owners automate marketing, content and daily operations through an AI assistant that can act, not just chat.

  • Action-taking assistant with confirmation cards before it publishes or spends credits
  • MCP server with OAuth so Claude and other AI tools can connect to a business account safely
  • Business memory, voice notes with speech-to-text, and daily morning summaries
  • AI articles with SEO scoring and one-click CMS publishing, bulk email/SMS and subscription billing
  • Tenant isolation and prompt-injection defenses, shipped in containers with automated tests
MCPOAuthLLMsPythonPostgreSQLDocker
Agentic commerce · GEOOptimaGeeks · Sep – Nov 2026

Agentic e-commerce platform & AI discovery engine

A pre-launch store where personal AI assistants (ChatGPT, Claude, Gemini) browse products and sign up for membership on behalf of their users.

  • MCP server that lets AI assistants register users, check status and browse the real catalog
  • AI-readable site: llms.txt, Markdown page versions, JSON-LD structured data and IndexNow
  • API with PostgreSQL migrations, rate limiting, disposable-email blocking and a Mailgun outbox
  • Privacy-compliant deletion, staff CSV exports and end-to-end tests in GitHub Actions
MCPGEOPostgreSQLMailgunGitHub Actions
Private AI · RAG · Device mgmtOptimaGeeks · 2026

TripleX: private on-premise AI & POS fleet backend

Two systems for the TripleX A10 dual-screen POS: an AI assistant that runs fully on local hardware, and a cloud backend that manages merchant terminals.

  • Zero-cloud AI that answers questions across 150+ legal contracts with exact source citations
  • MCP link from the chat (LibreChat) to an internal database (NocoDB), plus a Telegram digest agent
  • Android Management API backend: kiosk lockdown, remote lock, reboot, confirmed wipe and QR enrollment
  • Device status polling, command history, constant-time API key checks and automated tests
RAGMCPLocal LLMsAndroid Management API
Full-stack platformTechLeadz · Client: Mozilla

Shepherd: campaign management for Mozilla Firefox Ad Ops

An internal platform that replaced error-prone spreadsheet workflows with one system for campaigns, line items, advertisers and inventory.

  • Owned the full lifecycle: architecture, development, testing, rollout and performance tuning
  • Django REST APIs and a React (Vite) frontend for every module
  • BigQuery analytics and Boostr CRM integration to automate ad-revenue reporting
  • Tuned views, queries and backend logic: processing dropped from 5 minutes to under 30 seconds

Adopted across Mozilla's Ad Ops organization, with data accuracy improved by 90%.

DjangoReactPostgreSQLBigQuery
Infrastructure · DatabasesTechLeadz · Client: Apple

Apple Hub Services: upgrades & zero-downtime migrations

Infrastructure and database work for Apple Hub Services' production systems on AWS and Kubernetes.

  • Led Amazon Linux 2023 upgrades across 8+ Kubernetes clusters, improving benchmarks by 35% and achieving SOC 2 compliance
  • Upgraded RDS PostgreSQL from v12 to v17 with zero downtime using blue-green deployments
  • Consolidated 5 standalone databases into one RDS instance with AWS DMS, cutting operational costs by 40%
  • Patched container packages in Kubernetes pods to close high-risk vulnerabilities
AWS RDSAWS DMSKubernetesPostgreSQL
02 / How I build agents

The guardrails are the product

The model gets you to a demo. These rules, which run through every agent above, get you to production.

01

Nothing is saved without a yes

The agent shows a preview first and writes only after a person approves it. Approving a search never counts as approving a change.

02

Every answer shows its sources

Each field carries where it came from, and anything unclear is flagged for checking instead of guessed.

03

Paid tools are opt-in

Free sources by default. Paid APIs run only on request or approval, with caps, and every answer lists the tools it used.

04

Least privilege, always

Minimal connector access, OAuth, strict tenant isolation, and defenses against prompt injection.

04 / Experience

Where I've worked

Jun 2026 – Present

Senior Software Engineer · OptimaGeeks

AI agents, MCP servers and Python backends for client products: AI marketing automation, agentic e-commerce, private on-premise AI and an AI fundraising agent.

Oct 2021 – May 2026

Senior Software Engineer · TechLeadz IT Solutions

Built Shepherd for Mozilla Firefox Ad Ops and ran infrastructure and database upgrades for Apple Hub Services. RESTful APIs, PostgreSQL performance, AWS and Kubernetes, and code review for a team of 5+ developers.

2017 – 2021

Web Developer · earlier roles in Lahore

PHP, Laravel, CodeIgniter and Vue.js applications, including a telehealth platform with Twilio video calls and an employee management system.

2014 – 2018

BS (Hons.) Computer Science · Government College University, Lahore

Flying Languages

flyinglanguages.com

NLP-based language practice platform with AWS Polly and Transcribe for speech. Django, Vue.js, Celery, Redis.

DJ & event management

eventvibe.com

Email automation, Instagram tracking and event workflows, with AWS SNS/SQS for bulk messaging. Node.js, React, Docker.

Doctors For Health

doctorsforhealth.co.uk

Telehealth appointment system with Twilio audio and video consultations. PHP CodeIgniter.

05 / Toolbox

What I work with

AI & LLMs

AI agentsMCPRAGClaudeGeminiChatGPTPrompt engineeringHuman-in-the-loop

Backend

PythonDjangoDjango REST FrameworkFastAPICeleryRedisNode.js

Data

PostgreSQLMySQLBigQueryMongoDBDebeziumKafka

Cloud & DevOps

AWS RDSAWS DMSEC2SNS / SQSDockerKubernetesArgo CDGitHub ActionsTerraform

Frontend

ReactViteNext.jsTypeScript

Practice

PytestTDDCode reviewSystem designPerformance tuningAgile / Scrum
06 / Contact

Hiring for senior backend or AI engineering? Let's talk.