I'm Abdul Azeem — a backend engineer with 4+ years shipping production Python systems for fintech, accounting, and AI / LLM products. I've delivered 7+ live backends at Kavion.ai (Django, FastAPI, AWS Lambda) and built Invoize — an AI Bookkeeping Copilot for UK accounting firms with 7 specialised agents — end-to-end, solo. Live at invoize.co.uk.
Featured project: Invoize — AI Bookkeeping Copilot for UK accounting firms. 7 agents handling reconciliation, VAT sanity checks, cashflow forecasting, and month-end close. Built solo (Django · Next.js · OpenAI · Mistral OCR · QuickBooks · Xero) — live at invoize.co.uk
Not just extraction — Invoize reconciles, categorises, dedupes, sanity-checks VAT, prioritises payments, projects cashflow, and runs month-end close. 7 specialised AI agents handling ~70% of a bookkeeper's manual work. Django / DRF backend, Next.js / TypeScript frontend, Mistral OCR + OpenAI GPT-4o extraction pipeline, native QuickBooks Online + Xero OAuth. Built solo, live at invoize.co.uk.
Fuzzy-matches extracted invoices against bank statement lines: amount ± 5% tolerance, supplier name variations, date windows. Flags unmatched items for review. The #1 monthly-close pain, automated.
RAG over past invoice-to-account assignments. New invoice arrives → agent suggests the category with a confidence score. "This looks like your last 12 utilities bills — code to 6100?" One-click accept.
Same invoice uploaded via email, upload, and a Dext migration? Fuzzy match on invoice # + amount + supplier + date. Auto-merges high-confidence duplicates. Never post the same bill twice.
Scores every unpaid supplier bill on due-date proximity + supplier criticality, buckets them into pay-now / can-wait / defer. GPT-flagged critical vendors (rent, HMRC, utilities) always surface first.
UK-specific rulebook (HMRC Notice 700, 741A) encoded. Flags wrong VAT codes, missing VAT numbers, reverse-charge issues on EU services, zero-rated misclassifications — before they hit the VAT return.
Combines unpaid invoices, historical outflows, and LLM-detected recurring bills into a rolling week-by-week cash-balance walk. Surfaces the exact week a client goes negative — before it happens.
The orchestrator that ties it all together: reconciliation → VAT check → variance analysis vs prior months → generates a close report. Bookkeeper reviews the 3-5 flagged items instead of doing the whole close. Month-end shrinks from days to hours.
Multi-page PDFs or images → vendor, header, line items, UK VAT codes, currency, tax breakdown. Structured JSON in ~5 seconds with per-field confidence scoring and inline edits before pushing.
Multi-page statements → structured transactions with account-number masking, running-balance validation, and CSV + OFX exports that import into QBO and Xero unchanged.
Native OAuth into both. QBO gets Bill (invoices) + Purchase (receipts) push with vendor auto-create and per-line split billing. Xero gets ACCPAY Bills with UK VAT tax-type mapping. Duplicate check on both.
One firm login controls all client accounts. Each end-client gets their own workspace, integrations, and fully isolated data. Row-level tenant isolation, Fernet-encrypted credentials, JWT auth.
Each client gets a dedicated forwarding address. Suppliers email invoices directly; we auto-triage receipt vs invoice vs statement, extract, and route to the right client workspace. No manual sorting.
Owner / Approver / Bookkeeper roles. Bookkeepers extract and prep, approvers sign off before the bill hits the ledger. Optional per-client approval gate for high-value ledgers. Full audit log.
Free tier is 50 pages, no card required — you can process real documents in under 60 seconds. Or ask for a code tour and architecture walkthrough.
An AI-first stack tuned for shipping production-grade agentic products. Python backend, modern LLM tooling, cloud-native infra.
Multi-tenant SaaS architecture. REST APIs, async task pipelines, OAuth flows, encrypted credential storage.
LLM-powered extraction, RAG pipelines, and agentic workflows. Hosted APIs and local inference both in production use.
Type-safe React UIs end-to-end. Dark-mode design systems, responsive layouts, accessible flows.
Containerized deploys, Fernet-encrypted secrets, CI pipelines, AWS-hosted production workloads.
I'm Abdul Azeem, a backend engineer based in India with 4+ years of production Python experience. Currently a backend developer at Kavion.ai (N-Labs AI Pvt Ltd), where I've shipped 7+ production backends for fintech, accounting, edtech, and HR-tech — handling SAP, Razorpay, PhonePe, WhatsApp Business, and Stripe integrations end-to-end.
I specialise in multi-tenant SaaS, LLM-powered pipelines, and OAuth-heavy integrations. On the AI side I've shipped production code with OpenAI, Claude (Anthropic SDK), Mistral OCR, LangGraph, pgvector, and RAG. On the infra side: AWS Lambda microservices, Docker, GitHub Actions, Postgres at scale, Celery / Redis async pipelines.
My most recent solo build is Invoize — an AI Bookkeeping Copilot for UK accounting firms, live at invoize.co.uk. Not just extraction: 7 specialised agents handle reconciliation, auto-categorization, duplicate detection, payment prioritisation, UK VAT sanity checks (HMRC 700/741A), cashflow forecasting, and month-end close orchestration. I designed the architecture, the LLM pipeline, the admin + client portals, the OAuth flows into QuickBooks and Xero, and shipped it end-to-end. It's the project I'm most proud of.
Currently open to senior backend / AI engineering roles — remote-first, India / UK / US time zones. Notice period: 30 days.
Looking for a backend engineer who can ship Python systems end-to-end and is genuinely comfortable across LLMs, OAuth integrations, and multi-tenant SaaS? Let's talk. Recruiter, founder, or hiring manager — happy to do a screen, code walkthrough, or live demo of Invoize.