Software Engineer — Backend Systems & Applied ML Islamabad, Pakistan · saadamir070@gmail.com · LinkedIn · Medium

I build backend systems in NestJS/TypeScript and, more recently, production ML pipelines in Python. Currently a Junior Data Scientist at Innovo Technologies, working on a last-mile delivery platform — the routing/dispatch engine on the backend side, and the demand forecasting models on the ML side. Previously spent a year building backend systems full-time (HR/LMS platforms, real estate tokenization, auth systems).

Innovo Technologies — Ground-up rebuild of a B2B last-mile delivery platform (NestJS API, React portal, PostgreSQL, Python/FastAPI). I own the routing engine, live vehicle tracking, dispatcher control tower, and territory/customer management modules. On the ML side, built and shipped a reorder-prediction model (XGBoost) with per-segment dynamic windows, now in production for demand planning; currently extending into zone- and SKU-level operational forecasting.

MicroAgility — Built APAC, an enterprise HR/LMS platform (NestJS, MongoDB, TypeScript): authentication, multi-stage approval workflows, leave management, document uploads via GCS, and paginated REST APIs.

NxtOf — Full-stack work on PropNerd, a real estate tokenization platform. Implemented passwordless auth with WebAuthn (FIDO2) and biometric login; built APIs with NestJS, GraphQL, and PostgreSQL.

Futurenostics — Gaming hub and e-commerce projects on the MERN stack, deployed with Docker/Kubernetes and GitHub Actions CI/CD.

NestJSTypeScriptNode.jsGraphQLFastAPIPostgreSQLMongoDBTypeORM — backend PythonXGBoostPyTorch — ML ReactNext.jsReact NativeTailwind — frontend DockerKubernetesAWSGitHub Actions — infra

Multi-Stage SaaS Backend — Multi-tenant NestJS backend with JWT auth (access + refresh), RBAC, GraphQL via Apollo, real-time chat/notifications over Socket.IO, and 90%+ test coverage. Deployed on Render.

MeriCloset — Fashion recommendation platform (MERN, TypeScript) with collaborative filtering, a Gemini-powered chatbot, Stripe payments, and a scraped catalog of 500+ products.

Multimodal RAG System — Retrieval-augmented generation over text, tables, and images, with multiple reasoning strategies and a built-in evaluation framework. Hybrid Consensus Chain — CLI blockchain in Go with hybrid PoW-BFT consensus, adaptive CAP-theorem trade-offs, node trust scoring, and a UTXO model backed by a Merkle forest; tolerates up to 1/3 Byzantine nodes. Raft Consensus KV Store — Distributed key-value store implementing the Raft consensus algorithm. VAE vs GAN Comparison — Comparison of generative models (VAE, GAN) on CIFAR-10. English-Urdu Neural Translation — Fine-tuned Transformer (MarianMT) for English-Urdu translation, with BLEU evaluation and beam-search decoding. JotDown — Minimal note-taking app, React + Express.

I Spent Weeks Trying to Beat XGBoost With a Neural Network. It Won Anyway. — a GRU vs. XGBoost writeup from a production forecasting problem at work.

• AWS Generative AI / Bedrock
• Meta Back-End Developer Specialization
• AWS Cloud Foundations
• AWS Cloud Web Application Builder

BS Computer Science, FAST-NUCES Islamabad — Graduated June 2025