Irfana Aslam

Offline AI production pipeline on Blender + Python. Prompt → script → casting → scene graph → cinematography → neural voiceover → headless render → edit, subtitle, score → multi-format export. 39 modules, 29 films produced, $0 API cost, CPU-only.

Retrieval-augmented support agent that crawls the site nightly, processes PDFs and markdown, retrieves via FAISS vector search and answers with source citations. Runs entirely on local models.

Parses PDF/DOCX resumes, extracts 300+ technical skills, runs ATS compatibility checks and semantic scoring, then generates cover letters, learning roadmaps and interview preparation via local LLMs.

Gamified Python practice platform with a real server-side judge that runs submissions against test cases under a hard-enforced execution timeout and a code-safety pre-check, a step-through debugger that traces real interpreter execution, an AI code reviewer, a skill tree, career roadmaps and mock interviews.

A seven-stage pipeline that takes a hand-drawn building sketch and returns a simulated building. It retrieves similar CAD drawings, estimates depth, reconstructs the form in 3D, voxelises it into a storey-by-storey model, writes an EnergyPlus thermal model and ranks the options on a Pareto front. 54 modules with a 28-check test suite. Each stage has a research backend and a dependency-free fallback, so it runs on any machine. On a 288-image, 12-category benchmark it scores 0.934 Precision@1 and 0.613 mAP, and carries five sketches end to end in 2.6 seconds each on CPU. Mesh validation caught and repaired broken geometry in 3 of those 5 reconstructions. The same building evaluated across five climates ranges from 64 to 123 kWh/m²·yr.

Pixel-level segmentation of hairline cracks for structural inspection. Trained from scratch on a synthetic labelled dataset; achieved 0.785 Dice / 0.658 IoU / 0.858 recall on a held-out test split, trained entirely on CPU in 36 minutes.

Classical CV pipeline, MOG2 background subtraction with contour filtering and a Euclidean-distance multi-object ID tracker. Processed 900 frames of 1280×720 traffic footage at 30fps with no training and no GPU.

Live e-commerce price scraping across 29 product categories with Pandas processing, six interactive Plotly views, a Value Score ranking system, auto-generated insights and CSV/Excel export.

Dynamic KPIs, Prophet-based forecasting, vaccination tracking and RSS news sentiment analysis with country comparison and export.