Two Shows + Consulting · YouTube

THE
AI
UNPACKED

Unpacking AI. Simplifying the Future.

AI judgment, not AI commentary — for the people deciding where the budget goes. Practical masterclasses, sharp enterprise analysis, and hands-on consulting from a 15-year practitioner.

▶  Watch on YouTube ➞  Work With Me
Tutor Tuesday

Tutor Tuesday

Every Tuesday · 7:00 PM IST

One concept. One live demo. One thing you can use tomorrow. A weekly AI masterclass you can deploy by Wednesday.
The AI Unpacked Show

The AI Unpacked Show

Every Thursday · 7:00 PM IST

The show AI leaders can't afford to skip. Tension · Twist · Takeaway — what the week's AI news means for your market.
Consulting

Work With Raj Jain

AI Consulting & Architecture

Strategy, GenAI solution design, and AI governance — on contract, for enterprises shipping AI into production.
🎓

The Raj Lens

AI Enthusiast — 15+ years across enterprise tech, cloud & consulting.

📊

Data-Grounded

Every claim anchored to a named, verified source with a specific number. No vibes.

🌎

Globally Local

Global AI news mapped onto the US, UK, Singapore & Australia markets.

⚡

Essential, Not Just Informational

Content that makes you feel you need to act — not just nice to know.

The Two Shows

Two Formats.
One Mission.

Tutor Tuesday
TUE · 7:00 PM IST

Tutor Tuesday

"One concept. One demo. One thing you can use tomorrow."

A weekly AI masterclass that turns abstract concepts into immediate, profit-driving action. One idea gets unpacked, demonstrated live, and handed over as something you can deploy by Wednesday morning — grounded enough for leaders, practical enough for a curious beginner.

Signature Episode Arc
Introduction→ Concept→ Live Demo→ Apply→ Takeaway
A grounded idea, a live proof, and a deployable asset — every episode
The Concept Card: one branded, screenshot-ready slide per idea
The Tuesday Takeaway: a prompt, setup guide, or framework you can use today
Speaks to beginners, practitioners & leaders in a single arc
▶  Watch Episodes
The AI Unpacked Show
THU · 7:00 PM IST

The AI Unpacked Show

"The show AI leaders can't afford to skip."

Not AI commentary — AI judgment. The internet is saturated with people reporting AI news. The opening is in reading what the news means for an enterprise that has to make a decision on Monday. A tight ~10-minute talking-head, every week.

The Format Engine
Introduction→ Tension→ Twist→ Takeaway→ Quiet Story→ Close
The Tension: one problem, mapped across the US, UK, SG & AU markets
The Twist: the contrarian read — the Raj Lens on what headlines miss
The Takeaway: three concrete moves you can test against your org today
The Quiet Story: the under-reported signal — the reward for staying
This Week · Week of June 3, 2026

"Your country is spending billions on AI. Most of it is being wasted." — $547B in US enterprise AI spend, zero measurable ROI.

▶  Watch The Show
Tutor Tuesday · Season One

12 Episodes.
4 Arcs.

Episodes group into thematic arcs so casual viewers become subscribers chasing a series — not stumbling on a single video. Every episode stands alone for new viewers, while the arc rewards bingeing.

Foundations
How AI Actually Works
How AI Thinks
Hallucinations
Context Engineering
The Operator
AI for Individual Performance
The LLM Council
One-Shot Prompting
Personal AI Workflows
The Business
AI for Profit & Teams
Automating the Boring 20%
AI for Sales & Outreach
Cost-Per-Task Analysis
The Edge
Frontier & Strategy
Meet the Agents
Where AI Is Heading
Building Your AI Moat
About the Host

Raj Jain

Raj Jain
✦ CISM (ISACA) · AWS Certified
✦ CSPO · CSM
✦ IIM Ahmedabad Executive Alumni
✦ 15+ years · India, USA & UK
✦ Enterprise AI Strategy & Platform Architect
✦ AI Educator & Content Creator

I help enterprises turn AI ambition into working systems — securely, compliantly, and at scale. Over 15 years I've designed and shipped cloud architecture, security frameworks, and AI-integrated products for startups, MNCs, and everything in between. The same practitioner lens runs through both shows and every consulting engagement: what would I actually tell a client who has to decide on Monday?

GenAI & LLM Integration

AWS Bedrock, RAG systems, agentic workflows built for production.

AI Solution Architecture

End-to-end on AWS — from data pipeline to deployment.

AI Governance

PDPA (Singapore), GDPR, ISO 27001, DPDP — responsible AI in practice.

Secure AI Architecture

Zero Trust, IAM, DevSecOps applied to AI workloads.

✦

Designed and deployed an LLM-powered document intelligence pipeline for a legal firm — reducing manual contract review time by 35%, with full PDPA-compliant data controls.

✦

Built "Knoewit" — a Generative AI EdTech platform using LLM-powered content generation, personalised learning workflows, and integrated AI security controls.

✦

Reduced risk exposure by 40% across AI and cloud workloads through Zero Trust architecture and encryption strategies for AI data pipelines.

Selected Work

Projects in Production.

Three AI platforms I've architected end-to-end — each one a working answer to the same question the show keeps asking: how do you ship AI into a regulated, high-stakes domain without the governance being an afterthought? Every system below is built on AWS, powered by Claude via Bedrock, and designed so that compliance and Responsible AI are enforced in the architecture, not bolted on. Click any project to open the detail.

01
Legal Intelligence · LegalTech
Vakil AI
A three-agent legal intelligence platform for Indian law firms — document review, contract redlining, and litigation deadline management, with privilege and compliance enforced at every layer.
+
Claude Sonnet 4 / Bedrock LangGraph FastAPI + Kong PostgreSQL 16 + TimescaleDB Pinecone Celery + Redis React 19 ECS Fargate AWS ap-south-1
📱 Product Design — Lawyer Dashboard
Vakil AI lawyer dashboard with live agent activity
🤖 Design & AI Features
  • Document Review Agent — ingests PDF/DOCX, OCRs Hindi & English, retrieves Indian statute context, and returns a structured risk report with clause-level citations.
  • Contract Drafting & Redlining Agent — assembles jurisdiction-specific drafts from the firm's clause library, then diffs counterparty documents and classifies each change Accept / Negotiate / Reject with case-law citations.
  • Litigation Docket Agent — polls the eCourts API daily, calculates limitation periods under the Limitation Act 1963, and escalates deadline alerts through an Associate→Partner chain.
  • Multi-tenant by design: row-level isolation, a dedicated KMS key, and a separate Pinecone namespace per firm.
🔄 Agentic Workflow & MLOps
  • Each agent is a LangGraph StateGraph with typed state, named nodes, and conditional edges — sharing one Bedrock client and a common Postgres schema.
  • Human-in-the-loop gates via interrupt_before: HIGH-risk findings and every REJECT decision pause until a partner writes approval to the checkpoint.
  • Every run is logged to an agent_runs table — input/output hashes, model version, latency, and per-matter cost attribution.
  • Full observability through LangSmith traces, pinned to the same region for data residency.
⚖️ Responsible AI
  • GDPR Article 22 enforced literally: no consequential decision is ever solely automated — the graph cannot proceed past the human gate without partner sign-off.
  • Explainability trail on every assessment: AI reasoning surfaced beside the approval button so the human reviews the basis, not just the verdict.
  • Attorney-client privilege honoured structurally — documents tagged privilege=true are excluded from all non-essential pipelines.
  • MEDIUM/LOW outputs are auto-published but clearly labelled AI recommendations, never legal advice.
🛡️ Zero Trust & Security
  • Deny-by-default: short-lived 15-min JWTs, RBAC per route, mandatory MFA for Partner+, and Postgres row-level security so application bugs can't cross tenant boundaries.
  • All data stores in private VPC subnets, no public endpoints, WAF with OWASP Top 10, secrets only in AWS Secrets Manager.
  • Tamper-evident audit trail — an append-only TimescaleDB hypertable with a SHA-256 hash chain and S3 Object Lock (WORM).
  • Bedrock DisableModelTraining: true on every call; privileged documents never leave the VPC except to the in-region endpoint.
PDPA · GDPR · DPDP — one architecture

All three regimes are satisfied by the same controls, implemented once at the right layer. Data residency pins India tenants to ap-south-1 and Singapore tenants to ap-southeast-1 with cross-region replication disabled. Consent-first purpose binding is checked in API middleware before any request proceeds. Right to erasure runs a cascading delete across PostgreSQL, S3, Pinecone, and Elasticsearch within 72 hours. PDPA cross-border (S.26) and DNC checks before every WhatsApp alert, GDPR Art. 30 processing records and Art. 35 DPIAs, and DPDP data-localisation are all baked into the schema.

02
EdTech · University AI Tutor
Knoewit
A full-curriculum AI study platform that gives every enrolled student a 24/7 tutor grounded entirely in their own university's material — across every subject, from day one.
+
Curriculum-wide RAG Hybrid Retrieval (Vector + BM25) Cross-Curriculum Knowledge Graph LLM Re-ranking React Native + Web LTI / LMS Integration OCR Ingestion
📱 Product Design — Student App (iOS / Android)
Knoewit screen
University SSO & consent
Knoewit screen
Cited tutor answer
Knoewit screen
Verifiable source card
🖥️ Product Design — Educator Console (Web)
Knoewit educator console
Cohort overview
Knoewit educator console
Generated-content review queue
Knoewit educator console
Flagged-answer reproducibility trace
🤖 Design & AI Features
  • Curriculum-wide chat tutor — RAG over every subject, every department, answering only from lecture notes, prescribed texts, lab manuals, and past papers, with inline citations ("Lecture 7, slide 12").
  • Per-student learner model — concept-level mastery, recurring misconceptions, and pace, unified across subjects so a calculus gap shows up when it resurfaces in physics.
  • Content generation engine — practice problems, worked solutions, flashcards, and mock tests styled after the university's real past papers, constrained to syllabus scope.
  • Educator console — ships at launch, giving faculty cohort insight, corpus management, and tutor-policy control.
🔄 Agentic Workflow & MLOps
  • Bulk ingestion at curriculum scale — parallel per-subject pipelines parse PDFs, DOCX, PPTX, and handwritten notes via OCR, auto-tagging by subject, unit, and week with a live coverage dashboard.
  • Multi-stage LLM pipeline — query rewriting and decomposition before retrieval; hybrid dense+keyword search partitioned per subject; LLM re-ranking after; generation strictly from retrieved context.
  • An LLM preprocessing pass extracts a per-subject concept graph and links them into a cross-curriculum knowledge graph that powers cross-subject tutoring.
  • Standing evaluation suite — curriculum-grounded accuracy tests, hallucination probes, refusal tests, and red-team scenarios run against every model, prompt, or pipeline change; faculty-flagged answers feed back in.
⚖️ Responsible AI
  • Documented framework aligned to the NIST AI RMF, OECD AI Principles, and the EU AI Act transparency obligations, on five principles.
  • Transparency: every answer is cited; the learner model is inspectable; students always know they're talking to an AI.
  • Human oversight: mastery estimates inform tutoring, never grading; assessment-grade content passes a faculty review queue before release.
  • Fairness: bias audits across language backgrounds and disciplines; WCAG 2.2 AA accessibility; each university gets a model card.
🛡️ Guardrails & Security
  • Three-layer guardrails — input (prompt-injection & jailbreak detection, including instructions hidden in uploaded images; self-harm redirection), generation (groundedness check; Socratic mode during assessments), and output (toxicity, PII-leak, citation-integrity screening).
  • Per-user context isolation enforced structurally so no student's data can surface in another's session, then verified by a PII scanner.
  • Encryption in transit (TLS 1.3) and at rest (AES-256), least-privilege RBAC, comprehensive audit logging, and regular penetration testing.
  • LLM inference sub-processors contractually bound to zero-retention processing of student content.
GDPR & PDPA — privacy by design

The university is data controller, the platform a processor under a DPA. Data minimisation: institutional identity, enrolled subjects, and interaction data only — no advertising IDs, no location tracking, no identifiable student data used to train foundation models. Data subject rights (access, export, correction, deletion) are exercised in-app, fulfilled within statutory timelines (one month GDPR / 30 days PDPA). Residency: EU data stays in the EU, Singapore/ASEAN data in-region, cross-border transfers under SCCs. DPIA per deployment, 72-hour breach response, and guardian-consent flows where students are minors.

03
Wellbeing · Three-Sided Platform
Spiritual Raj
A three-sided spiritual-growth ecosystem — seeker app, mentor console, superadmin — where the principle "AI advises, humans decide" is enforced as a cryptographic signature check at the database layer.
+
Claude / Bedrock React Native (Expo) Aurora PostgreSQL Neo4j (Soul Graph) OpenSearch Lambda + Fargate Cognito Ed25519 / KMS pyswisseph
📱 Product Design — Seeker App & Mentor Console
Spiritual Raj screen
Daily reading & mentor feed
Spiritual Raj screen
Adaptive Readiness — 3 signals
Spiritual Raj screen
Cryptographic attestation gate
🤖 Design & AI Features
  • Three client surfaces, one backend — a saffron-sienna seeker app, a parchment mentor console, and a superadmin console — so identity, consent, and policy never diverge between surfaces.
  • Integration analyzer: Claude reads consented journals and reflections for depth-of-integration markers; mentor co-pilot prepares session notes; answer refinement powers the retrieval pipeline.
  • Seek (RAG) — answers from the founder's actual teachings: yt-dlp + Whisper transcription, chunked embeddings, vector retrieval, Claude refining answers that cite back to source videos.
  • Astrology engine on Swiss Ephemeris (sidereal, Lahiri, Whole Sign) feeding daily readings, dasha timelines, and advancement windows from one computation core.
🔄 Agentic Workflow & MLOps
  • Adaptive Readiness Engine — a state transition requires server-side concurrence of three independent signals: weighted engagement telemetry, a governance-gated AI readiness assessment, and an astrologically-computed advancement window.
  • All three are evaluated inside a single locked transaction that consumes the attestation token atomically, writes an audit snapshot, and triggers Soul Graph recalculation.
  • Window and muhurta rules live in versioned configuration tables, not code — auditable and adjustable without deployment.
  • The ephemeris suite validates against a fixed known reference chart before any window output is trusted; IaC via CDK/Terraform with CodePipeline CI/CD.
⚖️ Responsible AI
  • AI output is structurally powerless — only a cryptographically signed, single-use, time-limited mentor-attestation token (Ed25519 / KMS) can effect a protected state change, validated at the persistence layer by a database trigger.
  • Purpose-specific consent is verified at the moment of use, not merely at signup.
  • A distress-language guardrail runs before any assessment — routing distress to a human and suspending the readiness machinery for that seeker.
  • Every assessment carries a plain-language explanation and an immutable audit record; the approach is covered by a drafted 20-claim patent specification.
🛡️ Zero Trust & Security
  • Zero-trust by default — every request re-verified, least privilege, deny by default. Clients render state; they never compute eligibility — the server decides what is locked.
  • Entitlements enforced by server middleware on every gated route — a forged client tier claim does nothing.
  • Append-only schema: readiness assessments, phase transitions, billing events, and audit logs have UPDATE/DELETE trigger-blocked, so immutability survives bugs and privileged mistakes.
  • Payments behind signature-verified webhooks (Razorpay / Stripe); special-category data gated behind a separate explicit consent.
GDPR & PDPA — special-category data

The platform processes special-category data — beliefs, birth details, wellbeing — so trust is architectural rather than declarative. Residency-aware storage homes each member's data in their jurisdiction's region: India (DPDP), EU/UK (GDPR), Singapore (PDPA), and the US. Consent is granular, per-purpose, and unbundled, with special-category data behind a separate explicit yes. Binding mentor-confidentiality undertakings are recorded as proof before any sensitive-data access. Working data-subject rights (access, correction, erasure, withdrawal) and a breach register surfaced in the superadmin console complete the posture.

Work With Me

AI Consulting &
Architecture.

Beyond the channel, I take on a limited number of consulting and contract engagements. I don't just design systems — I understand what it takes to ship them under real constraints, justify them to a board, and make them stick after go-live. Founder background. Enterprise experience. Production mindset.

🚀

AI Strategy & Roadmap

Opportunity discovery, ROI modelling, and board-ready roadmaps that connect AI ambition to business outcomes.

🏗️

GenAI Solution Design

End-to-end architecture for LLM integrations, RAG systems, and agentic workflows on AWS — built for production.

🛡️

AI Governance & Security

Responsible AI frameworks covering prompt governance, bias auditing, and compliance with PDPA, GDPR & ISO 27001.

Open to conversations about
AI Consulting AI Architecture GenAI Solution Design Cloud Architecture (AWS) AI Governance Contract & Fractional Roles
Get in Touch

Let's Build
Something Real.

Whether it's a consulting conversation or a question about the show —

▶  Watch on YouTube ➞  Connect on LinkedIn 𝕏  Follow on X
📧  connect.real.raj.jain@gmail.com