EY India's cross-discipline technology challenge - teams build a responsible, production-minded solution to real partner business problems, pass AI and expert evaluation, receive mentorship, and pitch at an in-person finale.
Brabble editorial · Updated 15 Aug 2026 · Official page
Particulars
Organiser
EY India
Mode
Online build and reviews; in-person final jury
Prize / outcome
6.0: ₹3 lakh grand prize and ₹2 lakh runner-up award
Team size
Individual or up to 5
Application fee
Free
Rounds
4
What it is
EY Techathon is a national campus innovation challenge open beyond engineering. Techathon 6.0 focused on Agentic AI solutions for real problems contributed across pharma, BFSI, automotive, FMCG, retail, and IT-BPM. Public briefs define business context, agent roles, data assumptions, deliverables, and weighted criteria, so the challenge is less about inventing a flashy concept and more about executing a realistic workflow with guarded tools, synthetic or permitted data, measurable value, and a concise demo.
Rounds
01. Problem selection and project submission
Teams select an official industry problem, follow its prescribed deliverables, and submit architecture, agent/workflow design, assumptions, code/prototype, and a short live or recorded demo.
02. AI-enabled and expert screening
EY applies multi-stage evaluation using its tools and screening juries. Technical design, workflow realism, impact, UX, compliance, and demo criteria vary by problem statement and are often explicitly weighted.
03. Mentoring and semi-final presentations
Shortlisted teams refine their solutions with EY and industry mentors, address gaps, and defend progress before expert reviewers.
04. In-person grand finale
Finalists present live to the final jury; the jury score determines the grand prize and runner-up, while public voting supports a separate People's Choice award.
How to prepare
Step 01
Read the brief like a contract
Every required role, output, weight, and constraint should map to evidence in the submission.
Create a compliance matrix linking each deliverable and weighted criterion to a slide, demo moment, test, or repository path.
Select the problem where the team understands the workflow and can create realistic mock services/data, not merely the most familiar industry label.
Define what the agent may read, recommend, and execute; consequential actions need permissions, validation, logs, and human approval.
Step 02
Build an agentic system, not a scripted chat
The architecture must show useful decomposition and safe orchestration.
Give each agent a bounded role, typed tool contract, input/output schema, error policy, timeout/retry rule, and test cases.
Evaluate routing, retrieval, citations, task completion, latency, cost, and unsafe action rate against a fixed scenario set.
Use deterministic code for rules/calculation and LLMs where language/reasoning adds value; show fallback and abstention paths.
Step 03
Demonstrate workflow realism
A four-minute demo should prove the hardest business journey.
Use realistic synthetic records and mock APIs, including missing/contradictory data and a failed tool call.
Show behind-the-scenes agent activity without exposing hidden reasoning or secrets; make sources, approvals, and audit events visible.
Quantify one operational or commercial KPI, compare against a baseline, and disclose what still requires validation.
Representative practice - not leaked questions
Questions to practise
01
Build a loan-sales agent that can verify, underwrite, and draft a sanction letter without making an unauthorised credit decision.
What to show: Role boundaries, consent, deterministic policy, explainability, approval gates, audit, fairness, and adversarial prompts.
02
An RFP agent finds a document with malicious instructions embedded in it. What happens?
What to show: Prompt-injection isolation, untrusted-content labelling, tool permissions, source allowlists, output validation, and incident logging.
03
How do you prove a multi-agent workflow is better than one model call?
What to show: Baselines, ablation, task success, error containment, latency/cost, observability, and complexity justification.
04
The demo succeeds only on the happy path. Design the negative-test set.
What to show: Missing data, tool failure, conflicting sources, policy refusal, ambiguity, timeout, duplicate action, recovery, and human escalation.
Who can enter
Indian citizens residing in India, aged 18 or older, enrolled in a degree or diploma course in any discipline at an Indian college
Individual participation or teams of up to 5
Up to 2 non-students may join only where students remain the majority of the team under Techathon 6.0 rules
Free to enter; proof of identity, age, address, and student status may be requested
What you actually get
Techathon 6.0 awarded ₹3 lakh to the grand-prize team and ₹2 lakh to runner-up winner(s), plus a People's Choice award, mentorship, recognition, and potential exposure to real client challenges. The official terms define cash awards; general publicity about career exposure should not be converted into a promise of employment.
Frequently asked questions
Is EY Techathon only for engineering students?+−
No. Techathon 6.0 terms allowed students in degree or diploma programmes from any discipline at Indian colleges, with the stated citizenship, residency, and age requirements.
Can professionals join a student team?+−
Techathon 6.0 allowed up to two non-students in a team provided students remained the majority and total team size did not exceed five. Recheck this unusual rule each edition.
What makes an Agentic AI submission strong?+−
Clear role decomposition, realistic tools/data, safe action boundaries, resilient orchestration, measurable evaluation, workflow value, negative testing, and a demo that follows the exact business process - not merely multiple chatbots calling one another.
Should we use LangGraph, AutoGen, or CrewAI?+−
Use the framework that lets the team implement and test the required workflow reliably. Official problem pages may mention examples, but judges score design and outcomes; a simpler explicit state machine can beat an opaque framework graph.
Does reaching the finale guarantee an EY job?+−
No public Techathon 6.0 term guarantees employment. Treat mentorship, exposure, and project recognition as benefits separate from any later recruiting decision.
Checked against EY India's official Techathon 6.0 programme, newsroom release, official problem statements, and terms, 15 Aug 2026. Theme and technical criteria change substantially by edition; the detailed preparation above follows the published Agentic AI briefs and should be adapted when a new theme appears.