Google Gemini Hackathon 2026 Submission·Try it live
growpad.dev/pipeline/run-42
Pipeline Run #42
2m 47s · evidence-bundle-v3
Running
Progress5/8
Ingest
Synthesize
PRD
Tickets
Implement
Verify
Self-Heal
Export
$ growpad run --evidence ./uploads
✓ Evidence validated (4 files, 2.4s)
✓ Synthesis complete — 3 themes ranked
✓ PRD generated (12 reqs, 8 criteria)
✓ 8 tickets created with estimates
Generating implementation patch...

One pipeline that goes

It ingests evidence, writes PRDs, creates tickets, implements patches, and verifies its own output before shipping.

Try the demo
Pipeline Stages
Artifact Outputs
< minEvidence → Ship
Self-Heal Retries

Built for teams using modern stacks

Re
React
TS
TypeScript
Py
Python
Gm
Gemini
GH
GitHub
Pt
pytest
Nx
Next.js
Nd
Node.js
Re
React
TS
TypeScript
Py
Python
Gm
Gemini
GH
GitHub
Pt
pytest
Nx
Next.js
Nd
Node.js

How works

Three stages. One pipeline. Verified output every time.

Step 01

Upload interviews, support tickets, usage metrics, and competitor research. The pipeline parses every file format automatically and validates the bundle against schemas before processing.

user_interview_1.md2.4 KB
support_tickets.csv8.7 KB
usage_metrics.csv1.2 KB
competitors.md1.5 KB
Step 02

Synthesize & Decide

Gemini 3 clusters evidence across all sources, ranks themes by frequency × severity × effort, and selects the highest-priority feature with full rationale and confidence scores.

Users abandon onboarding — no progress indicator92%
Mobile experience breaks at 375px viewport71%
Search feature most requested in 47 tickets65%
Step 03

Generate PRD

Produce a complete product requirements document with overview, problem statement, acceptance criteria, and constraints — all linked back to source evidence. Fully editable before proceeding.

# PRD: Smart Onboarding Tracker
## Acceptance Criteria
Progress bar shows % and labels
Step validation before "Next"
Progress persists across sessions

8-stage

Scroll to follow each stage. The pipeline tests, detects failures, and self-heals.

Stage 1

Ingest Evidence

Parse interviews, tickets, metrics & competitor data into validated bundles.

Stage 2

Synthesize & Decide

Cluster evidence, rank themes, select the highest-impact feature with rationale.

Stage 3

Generate PRD

Produce full requirements doc with acceptance criteria linked to evidence.

Stage 4

Generate Tickets

Break PRD into engineering tickets with estimates, risks & file expectations.

Stage 5

Implement Patch

Generate unified diff from tickets and apply to target repo context.

Stage 6

Verify Tests

Run full test suite. Exit 0 = pass. Non-zero = trigger self-heal loop.

Stage 7

Self-Heal (≤2 retries)

Analyze failure, produce minimal correction patch, re-run verification.

Stage 8

Export Artifacts

Package PRD, tickets, evidence map, diff, test report & logs into zip.

Every design decision optimizes for trust and traceability.

Every pipeline run produces tests that verify its own output. If tests fail, the system self-heals with bounded retries instead of shipping broken code. You get proof, not promises.

No hallucination. Every PRD requirement, every ticket, every line of code traces back to source evidence with confidence scores. The evidence map is auditable and exportable.

PMs upload evidence and review PRDs. Engineers review diffs and test results. The pipeline bridges the gap with structured artifacts both sides understand.

Leverages Gemini 3's extended context window and reasoning capabilities for evidence synthesis, code generation, and self-healing. Open source, hackable, and built to last.

Every failed verification teaches the system. Run logs capture every decision, enabling post-mortem analysis and future fine-tuning. The pipeline gets smarter with each execution.

Everything you need to

Six pipeline stages. Zero guesswork. Full traceability.

Stop guessing.
Start verifying.

Ship features backed by evidence, tested by AI, and verified before merge.