Python/Flask developer building AI-powered backend systems that automate real-world business workflows. I like the messy parts: webhooks that retry, models that rate-limit, and the edge cases that only show up in production.

Currently building and running HandleHQ, a deployed multi-tenant SaaS that replaces manual customer support handling for small and mid-sized businesses using LLMs and REST APIs.

🎯 Exploring Solutions Engineering, Implementation, and Associate PM roles where hands-on technical skills meet customer-facing problem solving. US citizen, open to remote.

HandleHQ is a deployed, multi-tenant SaaS that answers a business's inbound customer messages with an AI agent, collects the details needed for a support ticket or sales lead, files the ticket, and hands the owner a portal to run the whole operation. There's a live demo, no login required.

What's actually in it:

• AI intake pipeline - OpenAI gpt-4o for strict-schema field extraction and gpt-4o-mini for conversational replies, off-topic classification, and ticket summaries, driven by per-business intake templates across 18 verticals.
• Multi-channel intake - email live via SendGrid Inbound Parse, a direct message API, and a WhatsApp channel (Meta Cloud API, built and pending Meta verification).
• Full ticket system - status, priority, tags, assignee, outcome, urgent flag, internal notes, and AI-generated summaries, with automatic escalation of overdue tickets up the priority tiers.
• Analytics - a portal metrics dashboard (ticket volume, average first-response time, ROI) plus an admin funnel view across all tenants.
• Multi-tenant data layer - PostgreSQL via SQLAlchemy with Alembic migrations, Redis for cross-worker rate limiting and portal login lockout, and nightly database backups with a tested restore path.
• Self-serve onboarding and billing - a public application form, admin review, Stripe Checkout (setup fee plus subscription with a free trial) behind a swappable payment-provider abstraction, and automatic tenant provisioning on payment.
• Three surfaces - a marketing site, a React/Vite/Tailwind customer portal, and a super-admin dashboard, each on its own subdomain.
• Production infrastructure - gunicorn behind nginx with Let's Encrypt TLS on a VPS, GitHub Actions CI running a hermetic pytest suite, Dependabot, fail-closed HMAC-verified webhooks, CSP/HSTS, request-ID tracing, and opt-in structured JSON logging.

The problem it solves: most small businesses handle customer inquiries manually with no structure or tracking. HandleHQ slots into that workflow and automates the intake, triage, and follow-up layers end to end.

• Reliability over cleverness - bounded retries, idempotency, graceful fallbacks: the stuff that keeps a webhook from taking down a worker.
• Integrations that hold up - most of my work is wiring third-party APIs together and handling what happens when one of them misbehaves.
• Explaining the system, not just shipping it - clear READMEs, architecture docs, and runbooks so the next person (or the customer) isn't lost.

Small, focused, MIT-licensed repos, most extracted and generalized from HandleHQ's production code. All tested.

• openai-retry - production-grade retry/backoff for the OpenAI SDK; treats insufficient_quota as terminal via a typed error, with jittered exponential backoff.
• idempotency-keys - framework-agnostic helper that runs an operation at most once per key; handles the in-flight conflict, never caches failures, memory or Redis backends.
• webhook-inspector - tiny Flask service to receive, verify (constant-time HMAC), inspect, and replay webhooks while you debug an integration.
• flask-error-envelope - consistent, typed JSON error responses for Flask APIs from a single call; never leaks stack traces, and sets an RFC-7231 Retry-After on 429s.
• sendgrid-inbound-parse-starter
  • minimal, documented Flask starter for receiving inbound email via SendGrid Inbound Parse.
• postman-to-markdown - a CLI that turns a Postman collection into clean, committable Markdown API docs.
• api-integration-demo - a small end-to-end demo relaying a form or JSON intake to Slack or Discord, with canonical field mapping, pluggable platform adapters, and bounded retry with backoff.
• rest-paginate - one iterator over any paginated REST API (page-number, offset, cursor, or RFC 5988 Link header); dependency-free with an injectable transport, a flexible items selector, and a hard infinite-loop guard.
• token-bucket - a dependency-free token-bucket rate limiter that paces outbound calls to stay under an API's limit; injectable clock and sleep for deterministic tests, thread-safe, with blocking and non-blocking acquire.
• openapi-diff - catches breaking API changes before they ship: compares two OpenAPI specs, resolves $refs and allOf, and exits non-zero in CI when a change would break existing clients.
• strict-env - fail-fast typed environment-variable config for Python: declare a schema, load it once, and every missing or malformed variable is reported at startup instead of one at a time; strict coercion, secret masking, and an injectable environ for testable config.
• circuit-breaker - a dependency-free, thread-safe circuit breaker that trips open after repeated failures, fails fast during an outage, then probes for recovery; injectable clock for deterministic tests and a generic state-change hook.

Before writing code full-time I spent 4 years managing operations at a medical clinic in New York. Promoted to manager within the first year, digitized the entire workflow, and introduced AI tools that cut administrative overhead significantly. That experience shapes how I work: the problem first, the technology second.

• Product: handlehq.net · live demo
• Email: saifullah4khan@gmail.com
• LinkedIn: linkedin.com/in/saifullah4khan