Applied AI
The applied half of a modern M.Tech AI syllabus, mapped to NITJ's AI-502 Computer Vision & NLP, AI-603 Social Network Analysis, AI-604 AI Video Analytics, AI-607 Medical AI and AI-632/634 Deep Learning for Signals & Images. Classical image formation and features, n-gram language models and tagging, video motion and behaviour, graph measures and influence, medical 2-D/3-D imaging and the IoMT, then digital signal processing and deep learning for signals. Same house style throughout: intuition first, a visualization you can play with, then the math.
Classical computer vision
How a 3-D world becomes pixels, how distinctive points and descriptors are found and matched, and how objects are detected and tracked across frames — the geometry-first foundation of AI-502 Computer Vision.
- Image formation and the pinhole cameraHow a 3-D scene becomes a 2-D array of pixels — the projection, focal length and field of view that every vision system has to invert. 10 min · intro
- Feature detection and matchingCorners, scale-invariant descriptors, ratio-test matching and RANSAC — the pipeline that gets from two images to a geometric transformation. 13 min · core
- Object detection and trackingFrom scored boxes to real object trajectories: IoU, non-maximum suppression, mean average precision, and tracking-by-detection with Kalman filters and data association. 14 min · core
- From handcrafted to learned featuresWhy SIFT, HOG, Haar and LBP were engineered, what a convolution layer learns instead, and when the handcrafted pipeline still wins. 13 min · core
Classical NLP
Language as a stochastic sequence: n-gram language models, smoothing and perplexity, then sequence labelling with Naive Bayes and the tagging/parsing tasks that precede neural NLP — the probabilistic core of AI-502.
- Language models and n-gramsA language model assigns a probability to a sentence. The n-gram approximation, maximum-likelihood estimation, smoothing and perplexity are where every modern model still begins. 13 min · core
- Tagging, parsing and Naive BayesSequence labelling with hidden Markov models, the Naive Bayes classifier, and the precision/recall confusion matrix used to evaluate both. 13 min · core
Video analytics
Video as a signal over time: background modelling and motion detection, temporal activity and behaviour recognition, and biometrics from face and gait — the analytical core of AI-604.
- Motion and background modellingVideo is a signal over time. Frame differencing and running-average background models separate what moves from what stays still — with real failure modes. 12 min · core
- Activity and behaviour recognitionRecognising actions over time: per-frame scores, temporal windows, smoothing, and the two-stream / 3-D convolutional architectures that made video classification work. 13 min · core
- Face and gait recognitionTwo biometrics from video: face recognition by embedding faces into a metric space, and gait recognition from the way a silhouette moves. 14 min · advanced
Social network analysis
Graphs of people and interactions: centrality and network measures, influence diffusion and link prediction, and the semantic web's RDF/OWL knowledge graphs — the core of AI-603.
- Network measuresA social network is a graph. Degree, closeness, betweenness and eigenvector centrality answer four different questions about who matters. 13 min · core
- Influence and link predictionHow ideas spread through a network — independent cascade and linear threshold — and how the same graph structure predicts which missing links will form. 14 min · advanced
- The semantic web and ontologiesRDF triples, RDFS/OWL ontologies, SPARQL queries and inference — how knowledge graphs represent facts machines can reason over. 13 min · advanced
Medical AI
AI on medical images and physiological streams: 2-D and 3-D imaging volumes with windowing and segmentation, and wearables and the Internet of Medical Things with real-time alerting — the core of AI-607.
- Medical imaging in 2-D and 3-DCT, MRI and their neighbours: Hounsfield units, window/level display, tri-planar viewing, and the 2-D and 3-D networks that segment anatomy. 13 min · core
- Wearables and the IoMTContinuous physiological sensing — PPG, SpO2, accelerometry — and the Internet of Medical Things: cleaning noisy streams, alerting on thresholds and doing it safely at the edge. 13 min · core
Signals and images
Digital signal processing for AI: sampling, Fourier spectra and IIR filtering of ECG/EEG, then 1-D convolutional networks that learn features directly from waveforms — the core of AI-632/634.
- Signals and digital signal processingSampling, the Nyquist limit, Fourier spectra, convolution and IIR filtering — the toolkit that turns raw sensor waveforms into clean evidence. 14 min · core
- Deep learning for signalsLearning the filters instead of designing them: 1-D convolutions, dilated stacks, spectrograms and the evaluation pitfalls of time-series models. 14 min · advanced
Generative applications
Turning pre-trained generative models into working products: prompt engineering as an empirical craft, then calling text, image and multimodal APIs with real request/response schemas, token and cost accounting, retrieval grounding, caching and moderation — the engineering core of AI-623 Applied Generative AI & Prompt Engineering.
- Prompt engineering in practicePrompts are programs written in natural language. Role, demonstrations, instruction, output format and constraints move a frozen model's output — and evaluation tells you whether they moved it the right way. 14 min · core
- Building with generative APIsA pre-trained model served behind HTTP is a product ingredient: request and response schemas, token and cost arithmetic, rate limits and back-pressure, retrieval grounding, caching and moderation. 16 min · advanced