Data Science undergraduate at FAST-NUCES Islamabad building production-grade agentic AI, NLP, and data engineering systems — shipped across two AI-firm internships, freelance enterprise engagements, and from-scratch ML research. I focus on systems where AI moves past prediction into autonomous decision-making: multi-agent LLM orchestration, real-time ETL automation, and hand-built NLP architectures.
$ cat .profile
ROLE = Agentic AI & Data Science Engineer
EXP = 2 AI-firm internships + freelance data engineering
DOMAIN = Multi-Agent Systems | NLP/LLMs | Data Engineering | Deep Learning
STACK = LangChain/LangGraph | PyTorch | BigQuery | React | Django
OPEN_TO = Job Opportunities · Research Collaborations · Freelance AI/ML WorkLanguages & Core
AI · ML · Deep Learning
Web · Cloud · Tools
Agentic AI & LLM Specialty
| Domain | Depth | Details |
|---|---|---|
| Agentic AI & Multi-Agent Systems | ●●●●● | LangChain/LangGraph orchestration, autonomous decision pipelines, human-in-the-loop checkpointing |
| NLP & LLMs | ●●●●● | Transformers, RAG, FinBERT, hand-built attention primitives, sentiment/NER, from-scratch language modeling |
| Deep Learning & Computer Vision | ●●●●○ | PyTorch, CNNs/RNNs/ViTs, OpenCV, medical-imaging segmentation |
| Data Engineering & ETL | ●●●●○ | BigQuery, Meta Ads/CM360/SA360 pipelines, advanced SQL, Pandas, GCP |
| Full-Stack Development | ●●●○○ | Django, React, Node.js, React Native, Supabase, MySQL |
▶ Agent Time Machine — Time-Travel Debugging for AI Agents
Stop re-running failed agents from scratch. Branch from any step of a LangGraph or AutoGen trace, edit the state, and resume live.
| Aspect | Detail |
|---|---|
| Stack | Python (FastAPI) · TypeScript · React |
| What it does | Records full agent execution traces, then lets you fork from any recorded step, mutate state, and resume the run through a web UI |
| Impact | Removes the need to re-run an entire agent trace from step 0 just to debug one bad decision midway through |
| Repo | View |
▶ review-rag-pipeline — Transformers Built From Scratch
Understand why a review says what it says. A from-scratch Transformer encoder/decoder pipeline with retrieval-augmented generation.
| Aspect | Detail |
|---|---|
| Stack | Python · PyTorch · Jupyter · Docker |
| What it does | Classifies product-review sentiment, retrieves similar reviews via a learned embedding index, and generates grounded natural-language explanations |
| Impact | Every attention primitive is hand-implemented — no nn.Transformer |
| Repo | View |
▶ cardiorisk — Production ML for Cardiac Risk
Production-grade ML toolkit for cardiac risk prediction, served through a validated risk-assessment app.
| Aspect | Detail |
|---|---|
| Stack | Python · scikit-learn/XGBoost · Streamlit · Docker |
| What it does | Clustering, ensemble methods (Random Forest, XGBoost), and neural networks on the UCI Cleveland Heart Disease dataset, with SHAP explanations for every prediction |
| Impact | Clean architecture, tested, containerized, CI/CD-ready |
| Repo | View |
▶ Agentic Trading Advisor — Multi-Agent PSX Investment Platform
A full-stack AI investment platform for the Pakistan Stock Exchange.
| Aspect | Detail |
|---|---|
| Stack | Python/Flask · React/TypeScript · MySQL · LangGraph |
| What it does | 10 specialized AI agents automate real-time market scraping, NLP sentiment analysis, and technical indicators to deliver personalized portfolio strategies |
| Impact | Built for scale across a full agentic backend and React frontend |
| Repo | View |
▶ UrduNewsLM — Statistical Language Modeling for Urdu
A from-scratch statistical language modeling pipeline for Urdu news text — no external NLP libraries.
| Aspect | Detail |
|---|---|
| Stack | Pure Python |
| What it does | Custom normalization, tokenization, stemming, and lemmatization; n-gram models with Laplace smoothing and Jelinek-Mercer interpolation; constrained headline/article generation with greedy, temperature, top-k, and nucleus sampling |
| Impact | Built entirely without spaCy, NLTK, or any pretrained tokenizer |
| Repo | View |
▶ Health Garden — Pakistani-Native Habit & Health App
A health app pairing local-unit food/workout tracking with a permanent, non-punitive "garden" that grows as habits stick.
| Aspect | Detail |
|---|---|
| Stack | React Native · TypeScript · Supabase (PLpgSQL) |
| What it does | Condition-aware (diabetes/PCOS/joint-safe) tracking, offline-first, bilingual Urdu/English, with a cost-capped AI coach on premium plans |
| Impact | Full mobile app: local-first data model synced to a Postgres backend |
| Repo | View |
| Metric | Context |
|---|---|
| ~60% reduction in manual processing time | Multi-agent reasoning pipelines automating 3 core decision workflows — AIM Lab |
| 2 days → <4 hrs stakeholder review turnaround | Real-time web-scraping modules + React dashboards — AIM Lab |
| 88–92% test-set accuracy | CNN/RNN models across 3 task domains — Arch Technologies |
| ~35% faster training convergence | OpenCV preprocessing pipelines — Arch Technologies |
| ~12 hrs/week manual reporting eliminated | Production ETL pipelines (CM360, SA360, Meta Ads → BigQuery) — Freelance |
| ~91% accuracy | NLP-based NER system on 1,000+ patient records — Freelance |
AI and Data Engineering Consultant · Freelance (Remote) · Jan 2026 – Apr 2026
- Delivered production ETL pipelines for Google CM360, SA360, and Meta Ads API for 3 enterprise clients, automating millions of rows into BigQuery
- Designed real-time BigQuery integrations for SleekFlow (CRM) and Tatango (SMS)
- Built an NLP-based NER system for a HIMS processing 1,000+ patient records at ~91% accuracy
BigQueryETLNLPNERSQL
Head of Development, Data Science Society · FAST-NUCES · Aug 2025 – Present
- Lead a 6-member dev team delivering 4+ web platforms and AI showcases per semester
- Mentor junior members in Python, Django, and ML best practices
LeadershipDjangoPythonMentoring
Agentic AI Intern · AIM Lab · Jun 2025 – Aug 2025
- Architected multi-agent reasoning pipelines using LangChain and LangGraph, automating 3 core decision workflows
- Engineered real-time web-scraping modules feeding live data into agentic systems; built React.js dashboards
LangChainLangGraphMulti-AgentReact
AI and ML Intern · Arch Technologies · Jun 2025 – Jul 2025
- Trained CNN and RNN models in PyTorch across 3 task domains achieving 88–92% test-set accuracy
- Built OpenCV preprocessing pipelines improving training convergence speed
PyTorchCNNRNNOpenCV
Aug 2023 – Aug 2027 · Coursework: Parallel & Distributed Computing, Deep Learning, NLP, Data Mining, AI, DBMS, Advanced Statistics
Certifications
| Certification | Issuer | Date |
|---|---|---|
| AI Agents and Agentic AI in Python Specialization | Vanderbilt University / Coursera | Jul 2025 |
| Generative AI for Data Scientists Specialization | IBM / Coursera | Dec 2025 |
| Google Prompting Essentials Specialization | Google / Coursera | Jun 2025 |
| Developing AI Applications with Python and Flask | IBM / Coursera | Jun 2025 |
- 🎉 Merged PR #1 in Shahoud867/Movie-Community-Management-System
- 💪 Opened PR #1 in Shahoud867/Movie-Community-Management-System
learning:
- Advanced LangGraph patterns — stateful agents with human-in-the-loop checkpointing
- Parallel & distributed ML — multiprocessing, async pipelines, per-stage profiling
building:
- Agent Time Machine # branch/edit/resume debugger for LangGraph & AutoGen traces
- cardiorisk # containerized, SHAP-explained cardiac risk ML toolkit
open_to:
- Job opportunities in Agentic AI / applied ML / data engineering
- Research collaborations in NLP, LLMs, or multi-agent systems
- Freelance AI/ML and data pipeline work
Open to job opportunities, research collaborations, and freelance AI/ML projects.