Hetu / Academy
Training forward-deployed engineers
Hetu Academy is a 10-week, cohort-based program that trains forward-deployed engineers in causal reasoning, decision verification, calibration, and AI adoption. Shared foundations for every trainee, then a split into one of two tracks — because "verify one decision" and "get anything live at all" are different jobs, staffed differently, and Academy trains for both on purpose.
Foundations — required, all trainees
| Module | Covers | Length |
|---|---|---|
| 01 · Causal Reasoning Fundamentals | DAGs, confounders, Simpson's paradox, structural causal models — why pattern-matching in language isn't the same as establishing causality in operational data. | 1 wk |
| 02 · Decision Verification Architecture | Designing a WHY-traversal tree, writing hard gates (temporal ordering, magnitude consistency, sample-size floors), the three-tier escalation model. | 1 wk |
| 03 · Calibration & Confidence | What calibration actually means, Brier scores and reliability diagrams, translating a statistical interval into a label a non-technical approver can act on. | 1 wk |
| 04 · Audit & Governance Design | Immutable decision-log schemas, named-approver workflows, what a model risk committee or regulator actually asks to see. | 1 wk |
Choose a track after week 4
Deepens the foundations: writing production-grade traversal trees, structural causal models for a named vertical, and the audit schema a real risk committee will actually accept. Ends with a vertical elective (below).
Legacy-system reverse engineering (reading undocumented code and inferring the business rules inside it), agentic workflow architecture beyond decision-verification, and the change-management reality of getting anything new adopted inside a risk-averse org.
Track B exists because most of what blocks enterprise AI adoption isn't a decision-verification problem — it's a legacy system nobody understands, or a workflow that was never going to survive contact with compliance. Every Track B placement doubles as market research: the hard problems its FDEs hit are the pipeline for new Vertical Verification Packs.
Track A electives — choose one or two
NBFC and bank origination pipelines, credit decisioning, collections. Model risk committee expectations, RBI and SOX context.
Marketing spend allocation, attribution reconciliation, margin-aware decisioning — taught jointly with Niti's team, using Niti's own open template library as case material.
Prior-auth exceptions, claims triage, fraud-adjacent decisioning under regulatory review.
Trade sign-off, position-limit exceptions, execution-quality verification.
Track B electives — choose one or two
Reverse-engineering undocumented systems (mainframe, monolith, decades of institutional logic) into a spec an agent can safely act on.
Designing and shipping bespoke agent-driven workflows outside the decision-verification frame — the general "build it and run it" job.
Getting a genuinely new system adopted inside a compliance-heavy org — the non-technical half of why AI pilots stall.
Capstone & outcomes
| Track | Cost | Commitment |
|---|---|---|
| Sponsored | ₹0 tuition | 12–18 month bond staffing Consulting or Deployment engagements |
| Self-funded | ₹1.5L one-time | None — certificate, capstone portfolio, alumni job board |
Sponsored seats are limited per cohort and prioritised for partner-college applicants. Self-funded seats are open to anyone who clears the technical screen. First two cohorts run in partnership with two Indian engineering colleges — named once the first cohort is confirmed.
Explore
Design partners
Say which track you're interested in — Decision Verification or General AI Adoption — and whether you're applying sponsored or self-funded.