India Cohort · One Core · Four Studios · Four Capstones
The AI conversation at your company is happening.
These tracks put you in the room.
You don’t build the model. You decide what it does, how it’s tested, whether it ships, and what it must never touch. Pick your role below — every track shares one Common Core, then goes deep in your world, and ends with a capstone on your real use case.
Product Manager / PO (product company)
Run the AI feature review meeting instead of surviving it.
You've approved AI features you didn't fully understand and hoped engineering caught the problems. Nobody taught you how to define 'done' for something probabilistic.
You write the AI Spec, set the eval gate, and challenge guardrail design — the Spec Owner in the room.
PM / BA on client accounts (services company)
The client's AI questions now have answers. With your name on them.
RFPs now have AI sections. 'How would you evaluate accuracy?' cannot be answered with a slide about innovation. The AI conversation on your account is happening without you.
You walk into discovery with an 11-stage methodology, a 5-signal suitability scorecard, and a white-label assessment deck.
Project Manager (PMP / Prince2)
The ADLC is the chapter PMBOK forgot.
Your estimates break on AI projects because there's a phase — Eval — no stage-gate template accounts for. Teams demo fast, then stall for months, and it lands on your timeline.
You chair the go/no-go with data, with phase-wise Definition of Done your team signs.
Scrum Master / Delivery Lead
The ceremony-facilitator role is shrinking. The AI Delivery Lead role is growing. Same you, four weeks apart.
You facilitate ceremonies for work you can't interrogate. Velocity says green; the AI feature hasn't passed an eval and never will at this rate. You've seen the Naukri search results for your title.
You know which ADLC phase the team is in, you protect the eval gate under pressure, and you learn story-file sharding — the skill the top AI-dev framework named after your job.
Business Analyst
You'll run the discovery session that decides if it's even an AI problem.
You sit in AI scoping calls nodding at 'embeddings'. Your BRD template has no field for confidence thresholds, failure modes, or what happens when the model is uncertain.
You run the AI intake interview and produce the 3-lane process map engineering actually builds from.
QA Engineer / Test Lead
You can't unit-test a language model. Your instincts are right — here's the new playbook.
The same input produces different outputs on different runs. Your regression scripts don't apply, but you're still expected to say 'this is ready'.
You design eval programs: golden sets, validated LLM judges, POC gates — evidence-based go/no-go without ML expertise.
System Administrator / IT Ops
Stop closing tickets. Own the AI that closes tickets.
Your company is deploying agents trained on the runbooks you wrote — and nobody asked you what they must never touch. The industry's advice was 'learn Python'. This is the other path.
You convert YOUR runbook into the agent's spec and run the human-review queue instead of being displaced by it.
SRE / Platform Engineer
Leadership bought an AI SRE. Make it prove itself.
An agent posts confident root-cause hypotheses in your on-call channel, and nobody defined what 'it was right' means. Trust it blind or fight it — neither is engineering.
Golden sets from replayed incidents, a judge rubric for RCA quality, zero tolerance on false-benign, read-only enforced by IAM — the agent earns the channel or you have the data.
The Four Role Studios
Same Common Core. Then your studio goes deep.
Week 0–1 is shared: plain-English AI, the ADLC, the Specify doc, Evals 101, and the AI Triage decision tree. Weeks 2–4 are your world.
Spec Studio
Heavy on: Forward Deployed Studio (stages 1–6)
- Week 2 — Intake & the 3-lane process map. Run the AI intake interview, map AI acts / AI flags / AI stops, deliver a suitability verdict — and hunt the 5 planted flaws first.
- Week 3 — Architecture decisions you own + the Specify doc. Write the 5-field behavioral contract to 'Agent-ready' score, then defend it in a live Grill Skill red-team duel.
- Week 4 — Eval thresholds & the executive brief. Set the eval gate in EvalForge, then present the go/no-go to a skeptical VP (your cohort).
You leave with: A complete AI Spec for your real use case, reviewed by a real engineer — plus the client-facing assessment deck (services variant).
Delivery Studio
Heavy on: ADLC Application (phase gates + Grill Skill)
- Week 2 — ADLC ↔ your delivery framework. Map the 6 ADLC phases to PMBOK stage gates or Scrum ceremonies; define phase-wise Definition of Done — why 'model built' is not 'model shippable'.
- Week 3 — Estimation & the eval gate. Estimate probabilistic work, chair a simulated go/no-go review with real eval data from EvalForge.
- Week 4 — The 2 AM incident. Trace forensics in the Admin Console: diagnose which guardrail failed before the 9 AM incident review, and run the change process on an eval threshold.
You leave with: An AI project plan with phase gates + a recorded go/no-go review you chaired.
Ops Studio
Heavy on: Infra deep examples + Admin Console
- Week 2 — Choose your world & autopsy the ghost project. ITSM, SRE, NOC, SOC, Storage, FinOps, Pipeline RCA, or CI/CD — walk a completed 11-stage project in your own domain and answer WHY at every stage.
- Week 3 — Runbook-to-Agent on YOUR runbook. Convert a real runbook into an agent spec: read-only-by-IAM guardrails, untrusted-input discipline (logs and tickets are attacker-influenced), lanes marked.
- Week 4 — Replay-based evals & the asymmetric trust metric. Golden set = replayed incidents. One catastrophic error class gets a hard zero bar. The agent earns your incident channel or it doesn't ship.
You leave with: A working eval harness over your replayed incidents + an agent spec your platform team can implement.
Eval Studio
Heavy on: EvalForge (full surface: taxonomy → judges → experiments → observability)
- Week 2 — Golden sets that mean something. Typical, edge, adversarial, out-of-scope — build the 20-case set for your use case and sabotage-check a broken one first.
- Week 3 — LLM-as-judge, validated. Build the judge rubric, weight guardrail compliance highest, and validate the judge against human labels.
- Week 4 — Eval programs in production. Regression suites on every prompt change, drift monitoring, and the data-backed go/no-go recommendation.
You leave with: A complete eval program: golden set, validated judge, POC gate metrics, and a go/no-go call defended with data.
The Capstone — chosen by AI Triage, not by guesswork
What is your system’s raw material?
A procedure becomes an agent. Conversations become a grounded assistant. Labeled history becomes a model. Documents become an extraction pipeline. Your capstone runs on YOUR use case — every path names its zero-tolerance error before a single tool is chosen.
Runbook-to-Agent (Agentic)
Raw material: A repeatable procedure humans execute — runbook, SOP, ticket-resolution steps
You build: An agent spec + replay-based eval plan: what the agent does alone, what it drafts for approval, what it never touches.
Zero-tolerance bar: False 'resolved' / suppressed real incident — one failure is an automatic NO-GO.
Reference builds: SRE incident triage · ITSM auto-resolution · FinOps cost anomaly agent · NOC alert correlation
Conversational System
Raw material: Back-and-forth language — support chats, intake questions, status queries, FAQs
You build: A grounded, citation-mandatory assistant with a refusal scope and a full-context human escape hatch.
Zero-tolerance bar: A confident wrong answer with a fabricated citation — automatic NO-GO.
Reference builds: ITSM deflection assistant · Support assistant · Conversational intake · Runbook knowledge assistant
ML / Predictive
Raw material: Historical records WITH outcomes — tickets + resolutions, transactions + labels, alerts + dispositions
You build: A classifier/anomaly/forecast plan: label audit, harm-matched metrics, per-class floors, confidence-routed gray zone, drift plan.
Zero-tolerance bar: The harmful miss class (missed fraud, missed P1) gets its own bar, separate from average accuracy.
Reference builds: Ticket classification & routing · Storage anomaly detection · Change-risk prediction · Churn scoring
Document Intelligence (IDP)
Raw material: Documents in, structured data out — invoices, claims, KYC forms, contracts
You build: A field-tiered extraction pipeline with confidence-routed human review and per-extraction audit trail — run against the live IDP agent over MCP.
Zero-tolerance bar: A wrong Tier-1 value (money, identity, legal) committed without human review — automatic NO-GO.
Reference builds: Invoice extraction (live IDP tool) · Claims processing · Contract clause analysis · KYC intake
India Pricing · UPI & EMI available
One certificate they can’t Google their way to.
Every option ends with artifacts only you could have built — on your own use case.
Self-Paced Studio Pass
₹6,999₹9,999
Common Core + one Role Studio + one capstone + ghost projects + community
Cohort (flagship)
₹17,999₹24,999
Everything in self-paced + 4 live Friday sessions + red-team duels + defended capstone + verifiable certificate
Additional Capstone Pack
₹4,999₹5,999
Any additional capstone path, self-paced — same spine, half the time
Eval Bootcamp (standalone)
₹7,999
1-week EvalForge-only sprint for QA leads & SREs