AI Solutions Developer & Data Scientist
I build production AI systems — conversational agents, geospatial intelligence tools, enterprise dashboards, and mobile apps — with a focus on real-world impact. Based in Lahore, Pakistan.
I'm a Data Scientist and AI Solutions Developer, currently serving as Co-Founder & CTO at Rapide Technologies, where I lead the development of AI-powered products and enterprise solutions for clients across finance, logistics, healthcare, and retail.
My work spans the full spectrum of applied AI — from fine-tuning language models and building agentic systems to geospatial intelligence platforms and production mobile apps. I hold a BS in Data Science from FAST NUCES Lahore.
I'm particularly drawn to problems at the intersection of AI and domain expertise — where intelligent systems can meaningfully change outcomes for real people and organizations.
Live products I've co-founded and built from the ground up.
A voice-first personal growth journal. Speak your thoughts — Whisper transcribes them, Claude analyzes for psychological patterns, mood trends, and emotional breakthroughs, then surfaces personalized affirmations and weekly trend insights. Available as a web app and React Native mobile app with CI/CD on Vercel and EAS.
AI-guided trauma discovery and emotional healing platform. Uses adaptive questionnaires and GPT-powered pattern recognition to surface emotional root causes and generate personalized healing roadmaps — privacy-first by design.
Production systems built for organizations across finance, logistics, and retail. Client names withheld for confidentiality.
Enterprise HR analytics platform with AI-powered dashboards, predictive employee turnover modeling, conversational data querying, and automated report generation with PDF/Excel export.
Full-stack lead generation engine that discovers, enriches, scores, and pitches leads across five strategic buyer personas — pulling from Google Maps, YouTube, Instagram, TikTok, Reddit, and neural search. Claude writes a personalised outreach hook for every lead based on real business signals. Complete browser UI with scraper dashboard, CRM pipeline, analytics, and encrypted API key management.
Agentic pipeline that tracks upcoming trading card game product releases in real time — running live web searches via Tavily, extracting structured data with a multi-LLM fallback chain (Claude → GPT-4o → Gemini → Grok), and outputting validated, deduplicated CSV/JSON with release dates, product types, images, MSRP, and source attribution. Built for a platform serving the global hobby and gaming industry.
Strategic market intelligence platform for a national bank. Computes Local Deposit Estimates and Area Opportunity Scores across pilot cities using OpenStreetMap data, Voronoi spatial analysis, and Claude-powered AI narrative insights with confidence tiering and full source attribution.
Interactive market mapping platform for a retail manufacturer. Scores dealer effectiveness, detects market saturation zones, and generates AI-powered strategic expansion recommendations using scraped competitor data, grid-based opportunity scoring, and Voronoi spatial analysis.
Three integrated AI tools built for a major logistics company: a WhatsApp-native executive finance assistant with multilingual Urdu/English audio support, a customer self-service portal with real-time shipment tracking and KPI dashboards, and a Bayesian-scored cross-sell intelligence tool with GPT-generated rationale.
Conversational geospatial data tool that converts natural language queries into SQL, executes them on uploaded datasets via DuckDB, and renders interactive H3 hexagonal grid visualizations — no SQL knowledge required.
Full internal HR management system for Rapide Technologies with dual admin/employee interfaces, project profitability tracking, paysheet management, work hour logging, real-time notifications, and Chart.js dashboards.
The most detailed publicly available sub-district map of Lahore — 801 verified named localities as Voronoi influence-zone polygons, distributed in Shapefile and GeoJSON formats for open community use in research and urban planning.
Deep learning, NLP, and computer vision research from FAST NUCES and independent study.
Hybrid solver combining BiLSTM, BERT, and a Directed Acyclic Word Graph (DAWG) to predict missing letters — achieving near-optimal Hangman gameplay performance.
Conditional GAN that synthesizes architectural floorplan images from text descriptions, with FID and IS evaluation metrics and a full custom dataset pipeline.
ML pipeline using XGBoost and neural networks to predict earthquake magnitudes, with feature selection, hyperparameter tuning, and full regression evaluation metrics.
Medical image classifier distinguishing mitotic from normal cells using VGG-11, DenseNet, and a custom CNN with focal loss and hyperparameter optimization.
Document retrieval system for Urdu text combining BERT, TF-IDF, Word2Vec, and Doc2Vec embeddings with SVM-based ranking and query reformulation — one of few IR systems built natively for Urdu.
Complete coursework from 8 semesters of BS Data Science at FAST NUCES Lahore — a widely used open resource for junior students across Pakistan.
I'm open to consulting engagements, collaborations, and interesting problems in AI, data science, and product development.
fatimaazfar381@gmail.com