🧠 You found the neural network!
Three.js r128 · WebGL shaders · ~130 nodes · 1 draw call
abdullahshafique2019@gmail.com
now building
Seeking AI/ML Engineering roles — contract & full-time, Q2 2025 onwards

// AI Systems Engineer · Pakistan · UTC+5
DEPLOY PROD.

Backend AI systems that are measurable, production-aware, and deployed. Every project ships with evaluation metrics, failure handling, and real performance data — no demos without numbers.

Python LangChain LangGraph FastAPI Groq FAISS Docker Streamlit
See the projects → Let's talk ↓ CV
Architecture
HOW MY SYSTEMS
WORK.
AI Research Agent
LangChain · Groq Llama 3.3 70B · Streamlit
🤖
User Query
Natural language research topic
Search Agent
Tavily API5 live sourcesReal-time
LLM Synthesis
Groq Llama 3.3 70B distils all content
Report Agent
Structures findings into sections + refs
PDF Export
Downloadable branded report in <60s
Live Streamlit appLlama 3.3 70BTavily real-timefpdf2 export
PDF Knowledge Assistant
FastAPI · React 18 · FAISS · Groq
📄
Upload PDF
pdfplumber extracts text page by page
Chunk + Embed
800 chars150 overlap384-dim
FAISS Index
Local cosine similarity · no cloud needed
Groq LLM
Llama 3.1 8B · answer in <3s
Answer + Citations
Page-level sources · relevance scores
Recall@5: 0.91<3s answersReact + FastAPIVercel + HF Spaces
House Price Prediction API
FastAPI · XGBoost · Docker
🏠
Raw CSV Data
California housing dataset
ML Pipeline
XGBoostRandomForestLinear
Model Selection
Best RMSE via cross-validation
FastAPI Endpoint
Pydantic schemas · strict validation
JSON Response
Prediction + confidence interval
RMSE: 0.48210ms p95120 req/sDocker + Gunicorn
LLM Eval Framework
ROUGE · BERTScore · GitHub Actions CI
🧪
Test Suite
50 hand-crafted prompts + expected answers
LLM Under Test
Any OpenAI-compatible model endpoint
Metric Computation
ROUGE-LBERTScoreFaithfulness
Report + CI Alert
GitHub comment if score drops
Regression Guard
Caught 2 regressions before deployment
50 prompts4.2s avg test2 regressions caughtGitHub Actions
Work
THINGS I'VE BUILT
(WITH NUMBERS).
01 — AGENT
🚀 Live Demo ↗ GitHub ↗
AI Research Agent
Real-time search · Llama 3.3 70B · PDF export in <60s
Autonomous agent that searches 5 live sources via Tavily, synthesises findings with Groq Llama 3.3 70B, and generates a structured downloadable PDF report. No copy-pasting. No tab-switching.
Stack: LangChain · Groq · Tavily API · Streamlit · fpdf2
Flow: Search Agent → Summarizer chain → Report Agent → PDF Exporter
LangChainGroq Llama 3.3 TavilyStreamlitfpdf2
report_agent.py
1from langchain.agents import AgentExecutor
2from langchain_groq import ChatGroq
3from utils.pdf_exporter import export_pdf
4 
5# Groq -- fastest inference available
6llm = ChatGroq(model="llama-3.3-70b-versatile")
7 
8# 5 live sources -> synthesise -> PDF
9result = run_pipeline(query, llm, pdf=True)
10# -> report.pdf under 60s
02 — RAG
🚀 Live Demo ↗GitHub ↗
PDF Knowledge Assistant
recall@5: 0.91 · grounded: 87% · <3s answers
Upload any PDF and chat with it in plain English. Page-level citations, multi-document support, multi-turn memory. FAISS runs locally — no cloud database.
Deployed: React on Vercel + FastAPI on Hugging Face Spaces (Docker)
RAGReact + FastAPIFAISSGroqHF Spaces
03 — ML API
GitHub ↗
House Price Prediction API
RMSE: 0.48 · p95: 210ms · 120 req/s
End-to-end ML pipeline + FastAPI serving. Input validation, error handling, Dockerized for production.
XGBoost best RMSE · Pydantic schemas · Uvicorn+Gunicorn
FastAPIXGBoostML PipelineDocker
04 — EVALUATION
GitHub ↗
LLM Evaluation Framework
50 prompts · 2 regressions caught · 4.2s avg
Automated testing + metrics for LLM outputs. ROUGE-L, BERTScore, faithfulness checker with GitHub Actions CI.
Caught 2 regressions before deployment · hybrid metrics (fast + accurate)
EvaluationLLMCI/CDBERTScore
05 — MONITORING
GitHub ↗
Agent Monitoring Dashboard
5 agents tracked · 30s alert time · cost logged
Lightweight dashboard for agent runs, token usage, failures, and cost per run. FastAPI + HTMX stack.
SQLite · Grafana optional · Slack alert on crash
MonitoringObservabilityFastAPIHTMX
0
Distinct AI Systems
agent · rag · ml · eval · monitor
0%
Avg Retrieval Recall@5
over 220 test queries
0
CGPA (x100)
Riphah International University
0yr
Design Background
graphic designer to AI engineer
About
ENGINEER.
BUILDER.
MEASURABLE THINKER.

Abdullah Shafique

AI Systems Engineer · Pakistan · UTC+5

I'm an AI Engineer with a background not many share — I spent 6 years as a professional graphic designer before transitioning into software engineering. That shapes everything: I build systems that are both technically sound and genuinely usable.

Currently studying Software Engineering at Riphah International University (GPA 3.99 / 4.0, 2024–2028) while building production AI systems in parallel.

Every project includes evaluation metrics, failure handling, and real performance data. No demos without numbers. Open to AI engineering internships and junior roles.

🏅 AI Agent Developer — Vanderbilt University 🏅 Python Development — Microsoft 🏅 Adobe Graphic Designer 🎓 SE @ Riphah · GPA 3.99
// Timeline
2024 — present
AI Systems Engineer (self-directed)
5 systems shipped, 2 with live demos. Measurable outcomes on every project.
2024 — present
BSc Software Engineering — Riphah University
GPA 3.99 / 4.0 · Building production systems alongside coursework.
2018 — 2024
Professional Graphic Designer
6 years, Adobe-certified. Applies design thinking to AI products and interfaces.
🟩
Active builder
Daily commits · 5 active repos
View GitHub ↗

Contribution Activity

Last 90 days · GitHub

AI / ML Stack

PythonLangChainLangGraph GroqOpenAI APIAnthropic API FAISSChromaDB Scikit-learnXGBoost

Backend / DevOps

FastAPIStreamlitDocker GitHub ActionsHF SpacesVercel

Engineering Practices

Evaluation-driven devFailure handling Latency optimisationCI/CD for AI Pydantic validation

Design (6yr background)

Adobe SuiteUI/UX Thinking Visual SystemsBrand Design

Most Used Languages

Proof
TRUST SIGNALS
& VALIDATION.
⚡ Action item: Request LinkedIn recommendations from 2-3 peers this week. Even one real sentence from a named person transforms this section from self-reported to socially validated. Replace these placeholder cards once collected.
★★★★★

91% recall on the first proper evaluation run. The metrics-first approach made debugging trivial — we always knew exactly what changed.

A
[ Replace with real name ]
Colleague · LinkedIn
★★★★★

The design background shows — every system Abdullah ships is both technically correct and genuinely pleasant to use. That combination is rare.

B
[ Replace with real name ]
Peer Engineer · LinkedIn
★★★★★

The LLM eval framework caught two regressions before they shipped. That defensive engineering separates reliable teams from flaky ones.

C
[ Replace with real name ]
Engineering Lead · LinkedIn
🟩 Daily commits · 5 active repos
⭐
🍴
🚀 2 live demos deployed & public
Contact
LET'S BUILD
MEASURABLE
AI SYSTEMS.

Seeking remote AI/ML Engineering roles — contract & full-time, Q2 2025 onwards.
UTC+5 · overlaps EU mornings & US East afternoons · fluent English.

Available now
🕒 UTC+5 · EU/US overlap
📅 Book a 30-min call ⬇ Download CV
GitHub ↗ LinkedIn ↗ Email ↗ WhatsApp ↗