Generative AI & LLMs
- Anthropic Claude API
- OpenAI API
- LLM integration
- Prompt engineering & versioning
- Structured outputs
- Context engineering
- Model routing
- Prompt caching
- Token & cost optimisation
Muhammad Arslan TabishFull Stack AI Engineer
I build LLM products end to end — database to browser.
Full Stack AI Engineer with 2+ years of professional experience building LLM-powered products end to end — Python/FastAPI services and data models through to the React 19 / TypeScript interfaces I design and ship myself.
- Years professional
- 2+Years professional
- Shipped projects
- 9Shipped projects
- Python · React · LLM
- Full stackPython · React · LLM
- Notice period
- 30 daysNotice period
Skills
Agentic AI
- Agentic workflows
- Multi-stage AI pipelines
- Tool / function calling
- LangGraph
- LangChain
- MCP
- Agent memory
- Multi-agent orchestration
- Human-in-the-loop
RAG & retrieval
- RAG
- Embeddings
- Vector search
- Vector databases
- Semantic search
- Hybrid retrieval
- Reranking
- Chunking & indexing
- Grounded generation
AI reliability & evaluation
- Schema validation & retry
- Output guardrails
- Automated QA gates
- Eval harnesses
- Retrieval evaluation
- LLM observability
- Prompt / model evaluation
- Failure analysis
Computer vision
- OpenCV
- YOLO
- Object detection
- Object tracking
- Real-time video processing
AI integrations & automation
- Apify
- Airscale
- MillionVerifier
- Instantly
- Third-party APIs
- Webhooks
- Lead-enrichment pipelines
- Automated outreach workflows
- Data extraction & processing
Backend
- Python 3.12+
- FastAPI
- async / await
- asyncio
- Pydantic v2
- SQLAlchemy 2.0 async
- PostgreSQL
- WebSockets
- Background jobs
- REST APIs
- API authentication
Frontend
- React 18/19
- TypeScript
- Next.js
- Vite
- TanStack Start / Router / Query
- Tailwind CSS
- Radix UI / shadcn
- Framer Motion
- React Three Fiber
- React Hook Form
- Zod
DevOps & production
- Docker
- Vercel
- Railway
- AWS S3
- Cloudflare
- GitHub / GitLab
- CI/CD
- Production monitoring
- pytest
Projects
AI systems
Self-built products — designed, built and run by me.
AI SaaS · self-built
Cold Outreach Pipeline
AI lead-generation SaaS
- A 9-stage, status-driven, resumable pipeline (extract, enrich, AI filter, gate, email, verify, AI copy, QA, sync) where every lead advances only on a valid stage transition, so runs crash and resume without duplicating work or LLM spend.
- A defensive structured-output layer for all AI calls — fence stripping, JSON recovery, schema validation and retry with a corrective nudge before a typed ParseError — plus deterministic guardrails and an automated QA gate that blocks weak copy before it reaches a prospect.
- Apify, Airscale, MillionVerifier and Instantly integrated into a single enrichment-and-outreach flow with webhook-driven status sync and per-provider fallbacks.
- Multi-tenant isolation, BYOK API keys encrypted at rest with AES-256-GCM, Stripe checkout with plan limits, per-call token-cost attribution, and a zero-spend demo mode using a mock LLM behind a Python Protocol.
- Python
- FastAPI
- Claude
- PostgreSQL
- Stripe
Conversational AI · self-built
Dukandar AI
Voice-first conversational shopping agent
- A catalogue-aware agent that walks shoppers through a facet-driven dialogue (category → kind → colour → size → price) and mirrors its filters live onto the storefront grid.
- Claude slot extraction constrained to facet values that exist in the live catalogue, eliminating hallucinated and out-of-stock suggestions, with a deterministic keyword parser as a $0 fallback on missing key, timeout or malformed JSON — the conversation never breaks.
- Session engine, faceted search, a TTS proxy for voice output and the storefront front end on async SQLAlchemy 2.0 + PostgreSQL.
- FastAPI
- Claude
- SQLAlchemy 2.0
- React
Market analysis · self-built
GoldWatch
AI gold (XAUUSD) trading analyst
- A market-analysis assistant that computes market structure and indicators, renders charts, gathers macro context and asks Claude for a read — with one hard safety rail: it never places an order, it prints a full briefing (entry, stop, target, lot size, money at risk) and waits for the human.
- Architected to resist LLM over-eagerness: a deterministic scanner scores setups first and an API call is spent only above a configurable threshold, keeping cost near zero on quiet days.
- Python
- Claude
- Market data APIs
Source not public — walkthrough on request
Full-stack & frontend
Client delivery — design systems through to deployment.
Freelance project · marketing site
Khaayaal — Elder-Care Platform
Six-route marketing site for an elder-care service
- Complete six-route marketing site on TanStack Start + Router with React 19 and Tailwind v4, on a custom design-token system and reusable brand component library.
- Animated 3D hero with React Three Fiber, scene-based scroll animation and an interactive pricing calculator over tiered care plans; React Hook Form + Zod, deployed on Vercel.
- TanStack Start
- React 19
- Tailwind v4
- Radix UI
- Vercel
Freelance project · Figma to production
Leads In The Pipe
B2B agency website on a serverless API
- Figma design translated into a production single-page front end on Vite + React 18 + TypeScript with Tailwind design tokens and Framer Motion, backed by a FastAPI serverless API.
- Vitest + React Testing Library + vitest-axe, a WCAG 2.1 AA pass and SEO work (meta/OG, JSON-LD, Core Web Vitals).
- React 18
- TypeScript
- Framer Motion
- FastAPI
- Vercel
Professional work
Intagleo Systems — internal systems, client and product names withheld.
Experience
Jul 2024 — Present
Software Engineer (Full Stack / Python) @ Intagleo Systems
Lahore
- Build LLM-powered product features on the Anthropic Claude API — agentic workflows, RAG pipelines and prompt-engineered integrations — with structured outputs, schema validation and deterministic fallbacks so AI features fail safely in production.
- Ship secure, scalable REST APIs in Python FastAPI — authentication, Pydantic validation, background jobs and business logic — across several production applications, plus Django/DRF services with role-based permissions.
- Design and optimise data across PostgreSQL, MySQL and MongoDB — schema design, efficient queries, migrations and data integrity across systems.
- Deliver full-stack features with React / Next.js front ends against FastAPI services, and real-time WebSocket features for live chat and notifications.
- Automate internal engineering workflows (task routing, approvals, compliance reporting) in Python, and deliver in Agile teams with Git/GitHub code review, sprint planning and weekly client meetings.
Jun 2023 — Aug 2023
Software Developer Intern @ Wisdom Software House
Lahore
Built full-stack web applications with a Python FastAPI backend and optimised MySQL queries for performance.
Dec 2022 — Jan 2023
Software Developer Intern @ Devsloop
Lahore
Web development fundamentals, version control, debugging and team workflows.
Education & achievements
Degree
2020 — 2024
University of the Punjab, Lahore
BSc Computer Science
Four years of core computer science — data structures, algorithms, databases and systems. I was interning and shipping full-stack work before I graduated, so the degree and the production experience were built in parallel rather than one after the other.
- Lahore, Pakistan
- BSCS
AchievementUniversity level
Runner-Up
Battle Byte Programming Competition
Second place at Battle Byte, a university-level programming competition. Timed, and scored on working solutions rather than elegant ones — the same instinct that shows up in production work: get something correct and shipping first, then make it fast.
- Competitive programming
CertificationUdemy
Python for FastAPI: Advanced Backend Development
Udemy
Advanced FastAPI — async patterns, dependency injection, authentication and production project structure. It maps directly onto the FastAPI services I ship at Intagleo and the pipeline behind Cold Outreach Pipeline.
- FastAPI
- Async Python
CertificationUdemy
Python for Computer Vision and Deep Learning
Udemy
OpenCV, deep learning fundamentals and image and video pipelines — the grounding behind the computer-vision side of my skill set: object detection, object tracking and real-time video processing.
- OpenCV
- Deep learning
About
AI systems, not AI demos.
I build AI applications that put LLMs into real enterprise workflows — and keep them reliable in production.
What I’ve built
- Built and deployed Cold Outreach Pipeline, a personal AI product that turns prospect data into personalised outreach. Reduced LLM API costs by 60% through guardrails, token controls and more efficient API usage.
- Improved a Django email-sending workflow, cutting request times from 10–15 seconds to 5 seconds or less — a reduction of at least 50%.
- Designed digital signage layouts for national quick-service restaurant chains, translating brand design systems into data-bound menu board interfaces.
- Redesigned an entire Next.js application’s visual theme, creating a more polished, consistent experience across layouts and components.
What I do well
- Full stack development with Python, FastAPI, Django, React, Next.js and TypeScript.
- LLM integrations, RAG and structured outputs that connect AI to business workflows.
- Reliability through validation, retry logic, failure recovery and human review.
- Turning complex business requirements into maintainable software and clear interfaces.
- Based in
- Lahore, Pakistan
- Open to
- Remote or relocation
- Notice
- 30 days
- Languages
- English, Urdu, Punjabi
Get in touch
Let’s build something.
Open to AI engineering roles — remote or on site — and to contract work on LLM systems that need to survive production.
- Phone
- +92 300 0835300
- muhammad-arslan-tabish
- GitHub
- github.com/Arslan9272
- Availability
- Lahore · remote or relocation · 30 days notice