Software Engineer / AI / ML / Full-Stack
I build production AI applications: a live-deployed autonomous LLM agent, a multi-tenant code-RAG SaaS, an exam-prep platform with fine-tuned DistilBERT, and the ingestion layer of a healthcare compliance copilot. FAST NUCES Software Engineering, 2026.
What I work with
Built across solo projects, team builds, and a production internship — every line below is backed by something I've shipped.
RAG pipelines, fine-tuned transformers, autonomous LLM agents, vector search, evaluation harnesses.
Type-safe UIs, cross-platform mobile apps, real-time experiences with SSE and WebSockets.
Layered architectures, REST APIs, async job queues, microservices, streaming responses.
Relational, NoSQL, and vector stores. Schema design, async ORMs, migration management.
CI/CD pipelines, containerization, live deployments on multiple cloud platforms.
Cryptographic protocols, threat modeling, attack simulation, in-CI vulnerability detection.
Multi-tenant architecture, third-party auth, billing webhooks, async job processing.
Production code in TypeScript and Python. Comfortable in JavaScript, Java, and C++.
What I've built
Each project below is grounded in a real GitHub repository. Click through for code, READMEs, and (for CI/CD Guardian) a live deployment.
Autonomous AI agent for CI/CD pipelines, deployed live on Render. FastAPI service ingests pipeline signals from a GitHub Actions workflow and issues its own block/allow verdict per build. Groq Llama 3.3 70B provides LLM-driven root-cause analysis on failures. Enforces branch protection, ≥80% test coverage, CVE detection, in-CI secret scanning, and PR approvals via the GitHub API. Multi-channel notifications across Slack, Discord, Teams, Email, and GitHub Issues. 75-test pytest suite enforced in CI, PostgreSQL long-term memory.
Final-year project: cross-platform AI exam-prep for Pakistani students (FBISE matric/FSc boards, MDCAT/ECAT entry tests). Fine-tuned class-specific DistilBERT models for past-paper topic prediction, evaluated on precision/recall/F1. RAG pipeline over the FBISE textbook corpus using sentence-transformer embeddings in pgvector, with Groq Llama 3.3 70B generating syllabus-aligned MCQ, short, and long-form questions. Confidence × log-frequency scoring predicts next-year topics with fuzzy-match evaluation. OCR pipeline (PyTesseract + pdf2image) for scanned past papers. Layered FastAPI backend with JWT + bcrypt + RBAC; React Native + Expo client on Android, iOS, and web.
Solo multi-tenant SaaS that lets users chat with any GitHub repository, generate documentation, review pull requests, and score task complexity, all powered by a RAG pipeline over indexed code. Voyage AI voyage-code-3 embeddings (1024-dim) with per-repo faiss-node vector indexes; Groq Llama 3.3 70B streams responses over SSE. Custom evaluation harness with rubric primitives (mustContain, mustCite) for grading retrievals. TypeScript monorepo: Next.js 16 + shadcn/ui frontend, Express 4 API, Prisma + PostgreSQL, Redis + BullMQ async ingestion. Clerk SSO, tiered Stripe billing (FREE: 1 repo / 50 chats/month, PRO: unlimited).
Real-time E2E encrypted messaging and file-sharing platform with a zero-knowledge server (stores only ciphertext and metadata). As part of the cryptography sub-team, implemented the protocol layer client-side via the Web Crypto API: AES-256-GCM for confidentiality, ECDH P-256 key exchange with signed handshake and key-confirmation MAC, ECDSA P-256 signatures, and a per-message HKDF ratchet keyed on sequence number. Replay protection via random nonces and monotonic sequence numbers, server-enforced uniqueness. Safety-number fingerprint verification for MITM detection, plus Node demo scripts showing replay and MITM attacks against secured and unsecured variants.
6-person team build of an AI compliance assistant for medical-device regulatory documents (FDA, ISO 13485, EU MDR) using a microservices architecture. Frontend: Next.js 16 + React 19 + Firebase Auth; backends: Node/Express + MongoDB Atlas and Python/Flask + FAISS + sentence-transformers + Groq Llama 3.3.
Academic project: multi-horizon stock prediction (1–14 days) for AAPL, MSFT, and BTC-USD using ARIMA, LSTM, and Ensemble models. Interactive Plotly dashboards, SQLite persistence, and full evaluation with RMSE, MAE, and MAPE. Built to explore classical and deep time-series modeling end-to-end on real market data.
Where I've worked
Production code, Agile sprints, real third-party integrations.
Built and shipped features inside a production MERN application: React components, Express routes, MongoDB schemas, and integrations with third-party REST APIs. Two-week sprints, daily standups, code reviews. Comfortable working alongside designers and product on real customer-facing modules.
Credentials
Who I am
I'm a Software Engineering student at FAST NUCES Islamabad (graduating June 2026). I care about systems that actually work in production: clean separation of concerns, real evaluation harnesses, deployments people can poke at.
My headline projects: an autonomous LLM agent running live on the public internet (CI/CD Guardian), a solo multi-tenant SaaS with real billing and a custom RAG eval harness (DevScope AI), an exam-prep platform with fine-tuned DistilBERT models (PrepifyAI), and the document ingestion layer of a 6-person healthcare compliance copilot (MedLaw).
I write most of my backends in Python + FastAPI or Node + TypeScript, design databases around PostgreSQL with pgvector, deploy with Docker and GitHub Actions, and ship UIs in Next.js or React Native.
Three months as a MERN intern at RISETECH Pvt. Ltd. taught me what production code review actually feels like — two-week sprints, daily standups, real third-party APIs.
"Available immediately for full-time roles — onsite in Islamabad / Pakistan, or remote worldwide. Especially interested in AI/ML engineering, agentic systems, and full-stack SaaS."
Get in touch
Open to full-time roles, internships, and contract work. The fastest way to reach me is email — usually a reply within a few hours.