SAAI 8 Week Apprenticeship Program

Program Cost: INR 14999.00 (includes GST)

SAAI Influence Intelligence Apprenticeship

Advanced Applied AI Apprenticeship

Build an AI system that reads beyond sentiment.

An eight-week research and engineering apprenticeship where three learners transform an existing Twitter Mood Analyzer into an ethical Influence Intelligence System—capable of examining coordination signals, propaganda techniques, narrative patterns, and evidence-based response strategies.

Apply for the first cohort Explore the learning journey

8 weeks of applied work 3 apprentices in the first cohort 3 rotating roles 3 intelligence layers 1 deployed capstone

Influence Intelligence Stack Build mode

0 Twitter Mood AnalyzerExisting sentiment and mood foundation Base

1 Coordination IntelligenceOrganic, coordinated, transparent, or deceptive signals Sprint 1

2 Agenda & Propaganda IntelligenceNarratives, techniques, emotional levers, and evidence Sprint 2

3 Response IntelligenceEthical intervention, text generation, and image briefs Sprint 3

Human review remains central. The system reports observable signals and confidence—not hidden intent or unverified accusations.

The challenge

Social media analysis cannot stop at “positive” or “negative.”

Real public conversations are shaped by coordination, framing, repetition, emotional triggers, agenda-setting, and the way people respond. This apprenticeship teaches learners to research these forces and translate them into testable, defensible technology.

Measure the mood

Start with sentiment, emotion, engagement, and the direction of public conversation.

Study coordination

Look for timing, repetition, similarity, shared sources, bursts, and abnormal patterns.

Understand narratives

Identify framing, agenda-setting, propaganda techniques, and emotional influence mechanisms.

Design responsible responses

Choose when to monitor, contextualise, prebunk, debunk, de-escalate, or not amplify.

How the program works

Learn the concept. Prove the skill. Ship the layer.

Every module connects foundational learning to a measurable assessment, an applied sprint task, a working product artifact, and a defense. Adaptive sequencing changes what comes next based on the evidence produced by each learner.

01 · Learn

Lessons & knowledgeConcepts, examples, vocabulary, research foundations, and tool introductions.

02 · Assess

Evidence of understandingDiagnostics, scenarios, research tasks, dataset audits, and practical builds.

03 · Build

Sprint executionModels, taxonomies, RAG pipelines, APIs, tests, integrations, and documentation.

04 · Defend

Explain every decisionResearch choices, model trade-offs, false positives, limitations, and ethics.

The journey board

Eight weeks. Three rotations. One intelligence system.

The learning story progresses from understanding an existing system to researching new intelligence layers, deploying APIs, and defending a complete research-to-production capstone.

Foundation

Diagnose & orient

Explore the existing Mood Analyzer, map its architecture, and establish technical, research, and domain skill profiles.

Outcome: personalised starting path

Sprint 1 · Discover

Research coordination

Study organic and inorganic reach, astroturfing, repeated content, timing, networks, and behavioural signals.

Outcome: taxonomy + dataset plan

Sprint 1 · Deploy

Build Layer One

Train a coordination pipeline, expose it through FastAPI, test false positives, and integrate the first layer.

Outcome: coordination API

Sprint 2 · Discover

Decode influence

Research agenda-setting, framing, propaganda taxonomies, misinformation categories, and emotional levers.

Outcome: annotation framework

Sprint 2 · Deploy

Build Layer Two

Develop a multi-label classifier, narrative clustering, a research knowledge base, and grounded explanations.

Outcome: propaganda + RAG API

Sprint 3 · Discover

Research intervention

Study correction psychology, prebunking, debunking, reactance, counter-narratives, and non-amplification.

Outcome: response decision system

Sprint 3 · Deploy

Build Layer Three

Train an intervention selector and generate grounded text responses, image briefs, and human-review flags.

Outcome: response intelligence API

Capstone

Integrate & defend

Orchestrate all layers, red-team the system, document limitations, demonstrate the product, and complete a viva.

Outcome: deployed Influence Intelligence System

The SAAI learning loop

Not session-led. Evidence-led.

Learners do not wait for the next class. The portal gives them the next lesson, task, challenge, or correction based on their current sprint, role, mastery profile, and submission evidence.

Discover

Study the problem, existing research, available datasets, terminology, and measurable questions.

Define

Create taxonomies, labels, annotation rules, variables, expected outputs, and ethical boundaries.

Develop

Prepare data, build baselines, improve models, compare experiments, and serialize complete pipelines.

Integrate

Build schemas, services, endpoints, validations, tests, documentation, and parent-system integrations.

Defend

Explain decisions, demonstrate the API, test edge cases, acknowledge limitations, and answer viva questions.

Expandable curriculum

Open each stage of the apprenticeship.

The curriculum combines technical AI research, behavioural and communication research, API engineering, RAG, testing, documentation, and responsible product decisions.

M0 Orientation & DiagnosticsUnderstand the product, workflow, standards, and your individual starting profile. +

Begin with the system that already exists.

Learners explore the current Twitter Mood Analyzer, run it locally, read the repository, map its architecture, and understand how the new layers will extend it. Technical, research, and domain diagnostics then shape each learner’s starting path.

Module outcomeA working local environment, architecture map, apprenticeship agreement, and individual mastery profile.

Core lessons

Introduction to the apprenticeship
Existing Twitter Mood Analyzer
The final product vision
Sprints, roles, and submissions
GitHub and documentation standards
Responsible AI boundaries

Initial tasks

Run the application locally
Read the repository structure
Create an architecture diagram
Submit an expectations note
Complete code-of-conduct and responsible-use acknowledgements

Diagnostics

Python, Pandas, ML, NLP, APIs, Git, testing
Research questions, sources, datasets, citations
Marketing, PR, propaganda, psychology, public opinion

M1 Common Research & Technical FoundationsBuild the shared language needed for research, modelling, RAG, APIs, and responsible analysis. +

Convert a vague concern into a measurable problem.

Students learn to move from claims such as “this campaign looks fake” to research questions based on observable variables: textual similarity, posting-time proximity, hashtag overlap, source diversity, and account behaviour.

Mastery requirement70% conceptual, 70% practical, and 60% research-methodology mastery before Sprint 1.

Research foundations

Product questions to research questions
Claims, assumptions, and evidence
Correlation versus causation
Dataset discovery and auditing
Leakage, bias, sampling, and label quality

Technical foundations

Text cleaning and tokenisation
TF-IDF, embeddings, and classifiers
Single-label and multi-label modelling
Macro-F1, micro-F1, and thresholds
Complete pipeline serialization

System foundations

RAG, retrieval, reranking, and grounding
Main, schemas, services, and model loader
Pydantic validation and API errors
Bias, defamation risk, and human review
Safe output language and guardrails

S1 Sprint 1 · Authenticity & Coordination IntelligenceResearch and build the first layer: how naturally or artificially a conversation appears to be spreading. +

From “viral” to measurable coordination signals.

The team investigates organic reach, paid amplification, coordinated advocacy, astroturfing, automation, repeated wording, timing similarity, hashtag overlap, URL overlap, engagement bursts, and account-network patterns.

Sprint outcomeA defensible taxonomy, labelled data approach, trained coordination pipeline, tested API, and model card.

Technical track

Review coordination-detection approaches
Audit datasets and engineer features
Train baseline and improved models
Compare performance and calibration
Export coordination_pipeline.pkl

Domain track

Research organic and inorganic amplification
Study astroturfing, bots, and coordinated accounts
Define labels and annotation guidelines
Create examples, counterexamples, and edge cases
Build a human-validation checklist

Integration track

Convert taxonomy into API schemas
Build model and feature services
Add validation, testing, and documentation
Connect the endpoint to the parent system
Prepare a deployment demonstration

POST /api/v1/coordination/analyze

S2 Sprint 2 · Agenda & Propaganda IntelligenceIdentify narratives, propaganda techniques, emotional levers, and areas that require evidence verification. +

Move from sentiment labels to influence mechanisms.

The team studies agenda-setting, framing, repetition, scapegoating, fear appeals, false dilemmas, authority cues, loaded language, dehumanisation, misinformation, disinformation, malinformation, and narrative clustering.

Sprint outcomeA multi-label propaganda classifier, narrative-clustering prototype, research knowledge base, grounded explanations, and integrated API.

Technical track

Evaluate propaganda datasets
Design a multi-label pipeline
Compare TF-IDF and embedding models
Tune thresholds and evaluate per-label F1
Export propaganda_pipeline.pkl

Domain track

Create the propaganda taxonomy
Define techniques and annotation rules
Map techniques to emotional triggers
Create difficult counterexamples
Build the approved research knowledge base

Integration track

Define multi-label response schemas
Connect the model and knowledge base
Build a basic RAG explanation service
Add evidence, safety checks, and tests
Integrate with Sprint 1 output

POST /api/v1/propaganda/analyze

S3 Sprint 3 · Counter-Influence & Response IntelligenceChoose responsible interventions and generate grounded text and visual communication briefs. +

Not every harmful-looking post needs a direct reply.

The team researches belief persistence, confirmation bias, reactance, prebunking, debunking, fact-checking, contextualisation, counter-narratives, de-escalation, media literacy, monitoring, and non-amplification.

Sprint outcomeAn intervention-selection model, RAG-grounded response generator, sample text posts, image briefs, safety validation, and end-to-end tests.

Technical track

Define intervention labels and examples
Compare model and rule-based approaches
Evaluate recommendation accuracy
Develop confidence thresholds
Export intervention_selector.pkl

Domain track

Research correction psychology
Define intervention strategies
Create a response decision tree
Define tone and image-generation guidelines
Create ethical review rubrics

Integration track

Build the intervention API
Integrate prior sprint outputs
Generate grounded text responses
Generate educational image briefs
Add safety and structured-output validation

POST /api/v1/intervention/recommend

M5 Final Integration, Red Team & Product DefenseOrchestrate every layer, test the system under pressure, and defend the final product. +

Complete the journey from research to production.

The final week connects the three intelligence endpoints into a single orchestration service, standardises schemas, adds versioning and logging, tests bias and safety, red-teams the system, and prepares the final demonstration and individual viva.

Capstone outcomeA deployed end-to-end Influence Intelligence System with documentation, test evidence, limitations, and a defensible product story.

Integrate

Connect all three endpoints
Create one orchestration service
Standardise response schemas
Add model-version information

Test

Add logging and error handling
Run bias and safety evaluations
Create adversarial scenarios
Validate human-review triggers

Defend

Prepare technical documentation
Demonstrate the product
Explain trade-offs and limitations
Complete individual technical viva

POST /api/v1/influence-intelligence/analyze

Rotating responsibility

Every apprentice researches, builds, integrates, and defends.

Each sprint assigns one primary role to each learner. The roles rotate so that nobody graduates saying, “I only wrote the report,” or “I only built the endpoint.”

Technical Research & Model Development

Investigate papers and datasets, design experiments, train models, evaluate performance, export pipelines, and prepare model cards.

Domain & Behavioural Research

Research marketing, PR, influence, psychology, propaganda, taxonomy design, annotation rules, ethics, and human-validation cases.

Integration & API Engineering

Translate concepts into schemas and measurable inputs, build services and endpoints, add tests, and connect each layer to the product.

Rotation

Student A

Student B

Student C

Sprint 1

Technical research

Domain research

Integration & API

Sprint 2

Integration & API

Technical research

Domain research

Sprint 3

Domain research

Integration & API

Technical research

Adaptive assessment sequencing

The next task depends on what your work proves.

A single total mark does not define readiness. SAAI tracks research, domain, data, modelling, RAG, API, testing, integration, documentation, communication, ethics, and defense as separate mastery dimensions.

Illustrative learner profileUpdated from assessments and sprint evidence

Technical research

Domain analysis

API engineering

Research method

Communication

Below 60

Repair the foundation

Receive a remedial micro-lesson, guided example, smaller task, correction exercise, and resubmission path.

60–74

Correct and strengthen

Explain errors, revise methodology, work through additional examples, and complete a partially guided implementation.

75–89

Apply independently

Take on comparative experiments, integration challenges, edge-case analysis, and structured peer review.

90+

Lead and extend

Attempt architecture challenges, red-team tasks, benchmark creation, advanced papers, and peer mentoring.

Assessment system

You are evaluated on evidence, not attendance alone.

The program uses multiple assessment formats because research ability, technical ability, integration ability, and defense ability cannot be measured through quizzes alone.

Knowledge & research

Concept assessments, scenario questions, source evaluation, paper summaries, research-gap identification, and literature comparison.

Build & integration

Dataset audits, preprocessing pipelines, baseline models, improved models, APIs, tests, RAG services, and end-to-end integrations.

Defense & judgment

Model explanations, dataset defense, failure analysis, ethical scenarios, live demonstrations, peer review, and individual viva.

The final capstone

One endpoint. Four levels of understanding.

The final orchestration service combines mood analysis, coordination signals, propaganda-technique detection, intervention recommendations, sample written content, image briefs, confidence scores, limitations, and human-review requirements.

Integrated multi-layer API
Three trained or hybrid intelligence pipelines
Research reports, taxonomies, and model cards
Test suites, red-team evidence, and safety notes
Live product demonstration and technical defense

{
  "mood_analysis": { ... },
  "coordination_analysis": {
    "coordination_probability": 0.79,
    "supporting_signals": [ ... ]
  },
  "propaganda_analysis": {
    "techniques": [ ... ],
    "claim_verification_required": true
  },
  "recommended_intervention": { ... },
  "sample_text_content": [ ... ],
  "image_brief": { ... },
  "limitations": [ ... ],
  "human_review_required": true
}

Scientific and ethical guardrail

The system is designed to describe observable patterns and uncertainty. It must not claim to know a person’s hidden intention, affiliation, or guilt.

What you finish with

A portfolio built from research-to-production evidence.

Every learner completes all three roles and leaves with artifacts that can be inspected, tested, discussed, and defended.

Three research cycles

Coordination, propaganda, and intervention research—each converted into definitions, data, models, APIs, and limitations.

Research evidence

All three professional roles

Technical researcher, domain researcher, and integration engineer—supported by individual commits and task evidence.

Role breadth

API

Deployed product layers

Three specialist endpoints and one orchestration endpoint connected to an existing applied AI product.

Production evidence

Model and system cards

Training approach, evaluation, assumptions, bias risks, failure modes, intended use, and human-review requirements.

Responsible AI evidence

GitHub contribution history

Commits, pull requests, reviews, tests, documentation, and shared ownership of the final product.

Engineering evidence

Technical defense

Live demonstration, architecture explanation, research defense, failure analysis, and an individual viva.

Communication evidence

Program fit

Built for learners ready to work without waiting to be taught every step.

This program is a strong fit when you…

Already understand basic Python and are prepared to strengthen it
Want to learn AI through research, experiments, APIs, and product work
Can document your decisions and defend your conclusions
Are comfortable reading unfamiliar repositories and papers
Can collaborate across technical and non-technical roles
Accept feedback, revision, testing, and resubmission as part of the work

This is not a passive course.

There may not be a live class for every concept.
Tasks unlock according to progress, dependencies, and mastery.
Research quality matters as much as code execution.
A working endpoint is incomplete without tests and explanation.
Every learner must understand the full system, not only their assigned role.
Human, ethical, and scientific judgment remain part of the final evaluation.

First cohort · Limited to three apprentices

Build. Research. Deploy. Defend.

Apply to help transform the Twitter Mood Analyzer into a deeper Influence Intelligence System while building demonstrable capability across AI research, human behaviour, RAG, model pipelines, API engineering, responsible AI, and technical defense.

Start application Ask a question

Frequently asked questions

Before you apply.

Will there be regular live sessions?

The apprenticeship is designed around lessons, adaptive assessments, sprint tasks, reviews, and mentor checkpoints rather than a fixed schedule of daily classes. Live guidance can be used where judgment, blockers, reviews, or defense preparation require it.

Do all apprentices work on the same product?

Yes. The first cohort collaboratively extends the same Twitter Mood Analyzer. However, each learner receives role-specific tasks, individual assessments, peer-review responsibilities, adaptive follow-ups, and an individual viva.

Will I only work in the role assigned to me?

No. Roles rotate across the three sprints. Every apprentice leads technical research once, domain research once, and integration/API engineering once. Everyone must also understand and defend the complete system.

Is the final product only a machine-learning model?

No. The final system combines trained pipelines, domain taxonomies, RAG knowledge bases, structured APIs, tests, explanations, confidence scores, safety checks, response generation, and human-review controls.

How will my next task be decided?

Your current sprint, role, prerequisites, previous submissions, criterion-level rubric scores, skill profile, and mentor review determine whether you receive remediation, correction, independent implementation, an advanced challenge, or a leadership task.

Does the system identify people as propagandists?

No. The project is designed to report observable signals, techniques, confidence, limitations, and areas requiring verification. It must not claim to know hidden intent, affiliation, guilt, or control without reliable evidence and appropriate human review.

Buy Now

Syllabus and Module Index

Explore modules and expand each card to view lessons.

Buy Now

Modules will be published soon.