Building AI systems people can actually run, not just admire.
I partner with businesses to turn broken, time-consuming workflows into automated software systems. Operating as BoltTaha, I handle the entire process, from diagnosing exactly what your team needs, to writing the code and keeping the infrastructure running smoothly.
Systems shipped, in order
Basketball Highlight Detection
Built a basketball highlight-detection system for a sports client: scans raw game footage and auto-exports clips of made baskets. Uses a two-stage AI pipeline: a geometry-based detector flags candidate makes, then a dedicated verification model (trained on 13,000+ labeled images) confirms the ball was truly in the net, cutting false positives. Added a review dashboard so staff can scan flagged clips in a grid, loop-play each one, zoom on the hoop, and reject false positives in one click, turning minutes of manual review into seconds. Latest validation: 100% of real makes caught in a 50-clip test batch.
GTM Sales Intelligence Assistant
Gohar Textile Mills' sales team needed an analyst on call just to answer routine revenue questions. I architected and deployed a hybrid text-to-SQL engine (8 zero-LLM certified query patterns plus a validated Bedrock Sonnet fallback), so staff now get auditable, data-backed answers in plain English in under 500ms, with zero analyst time.
Automated Extraction Engine
Built a custom software tool for Gohar Textile Mills that reads PDF purchase orders and automatically types the data into Excel, eliminating hours of manual data entry from their daily operations.
GTM Sales Forecasting Engine
Gohar Textile Mills' planners had no visibility into upcoming revenue: production, cash, and stock decisions were made blind. I architected an end-to-end forecasting pipeline: SAP data ingestion, a fiscal-calendar-aware time series engine, and a backtested 3-model competition (XGBoost, Prophet, Seasonal-Naive) that gives planners a rolling 6-month forecast per category and customer, roughly halving forecast error versus a naive baseline.
OCR Preprocessing Pipeline
Built a production OCR preprocessing pipeline for a US legal records company, cleaning an estimated 20M+ scanned property documents (1962–2026). A custom 5-stage OpenCV pipeline (deskew via Otsu thresholding, artifact/border cleanup, denoising, illumination correction via Gaussian + CLAHE, and adaptive thresholding) separates clean text from noisy scans. Runs as a parallelized, resume-safe batch system on Azure, with per-file manifests so multi-million-document runs recover from interruptions without reprocessing.
QR Payment Verification System
Screenshot-based mobile payments are common but trivial to fake or reuse, and businesses had no reliable way to verify them beyond trusting a photo. Built a full-stack fraud-detection pipeline: Gemini vision extracts transaction details from each screenshot, business rules catch duplicate images, repeat transaction IDs, and payment fingerprints, and a weighted risk engine flags anything suspicious. Deployed to EC2 behind Nginx with PM2 auto-recovery and a live React admin dashboard for review.
Experience
AI Engineer Intern
- Architecting and deploying an enterprise-grade, localized RAG-based agentic system to automate and optimize raw SAP production logs and multi-million row sales data matrices.
- Engineering end-to-end industrial automation pipelines using Python to completely eliminate manual data entry, streamline shift operations, and bypass operational bottlenecks.
- Running open-source large language models (Llama 3) completely offline on an internal corporate server, ensuring zero data leaks and 100% data sovereignty.
- Developing high-performance data processing pipelines using Pandas and LangChain to dynamically generate executive summaries and mathematical metric dashboards for the General Manager and Board.
Co-Lead (Development & Projects)
Collaborate with the core team to plan, manage, and lead cloud-based projects and technical initiatives, strengthening leadership, project management, and technical skills across the AWS ecosystem.
- Overseeing development projects using AWS technologies.
- Mentoring students and helping them explore cloud computing.
- Coordinating with the core team to organize workshops and hands-on sessions.
- Encouraging innovation through real-world AWS project building.
AI & Fintech Systems Engineer (Contract)
Engineered a production-grade backtesting engine evaluating 1,435+ historical trades with outcome-agnostic strategy evaluation and HRR risk analysis.
- Advanced data architecture: optimized OHLCV data storage using Parquet format across 24K+ files, cutting data query latency by 40% versus standard CSV baselines.
- Pipeline automation: built high-throughput data ingestion pipelines using Pandas, NumPy, and live financial APIs to handle real-time asset allocation and automated portfolio optimization logic.
- Dashboard deployment: designed and shipped an interactive Streamlit Cloud dashboard featuring live performance metrics and integrated continuous data pipelines for client review.
Stack & certifications
Client success & education
What clients say
“Would always be willing to hire this freelancer. He is so great, quick and hard working.”
“Muhammad was great. Fast, very smart and flexible. Would love to keep working with him on upgrades.”
“Taha is an excellent person, attentive to instructions and my requests. He demonstrated impeccable professionalism, and the results of his work were very satisfactory. I recommend Taha and will call upon him for my future technical needs. Well done, Taha!”
“A wonderful man.”
“Great, good work.”