Skip to content
ACAnkit Chatriki
Contact
Back to my workFILE 01 OF 02
CAL HACKS 13.0 · TOP 20 OUT OF 150 TEAMS

Fern

Engineering change & supply chain impact analysis

Turns every engineering change request into a financial impact assessment (cost, schedule and supply risk) from PLM and ERP data, in about 60 seconds.

View on Devpost
ASSESSMENT TIME~60sPer engineering change, with auditable calculations
DELAY IDENTIFIED13wkSupplier-driven, aerospace scenario
POTENTIAL SAVINGS$180KQualification savings evaluated
RESULTTop 20Out of 150 teams, Cal Hacks hackathon (Hyphenate track)

  1. The problem

    Engineering changes quietly blow up budgets and schedules. Open POs, scrapped inventory, rework and supplier lead times don’t get priced in until after the change is approved.

  2. The solution

    A multi-agent system (Claude Agent SDK, MCP) reads the ECR from PLM, computes the BOM delta, and runs cost, timeline and Monte Carlo lead-time models against ERP data in about 60 seconds. Agents only extract and explain; code computes every number, each tagged with its evidence class.

  3. Real example

    An engineer files a change for a cracked motor-controller bracket. Fern prices it at +$415 per aircraft plus $142K–$156K one time, and flags that the only qualified supplier’s 14-week lead time puts the change on the line 13 weeks late, before the board approves it.

Watch the demo1:01 · DEMO WALKTHROUGH, CAPTIONED, NO AUDIO

01The problem

R&D teams make thousands of changes over a project's life. Each one starts as an engineering change request (ECR).

1 ProblemSomething doesn't work
2 ECR filedAn engineer files a fix
3 Board reviewEngineering assessment
4 Approve or denyThe board decides

The true cost of an ECR is opaque

VISIBLENew part priceECR
HIDDEN
  • Open purchase orders ERP
  • Inventory scrapped ERP
  • Rework PLM
  • New tooling PLM
  • New supply chain risk

That is why a financial impact assessment matters for every ECR.

02What we promise

01Speed

An impact assessment for every ECR, in about 60 seconds.

02Total visibility

A clear picture of the downstream effects.

03Traceability

Every source, calculation and prediction is auditable and labeled.

FactComputationAssumptionJudgment

03How it works

PLMAn engineer files the ECR
FernAssesses the change
ERPThe released ECO lands
Assessor agentResearch agentChaser agentFern agent

Lightweight by design: it plugs into a company's existing software and AI stack.

04The forecast pipeline

ExtractAgent reads the ECR
BOM deltaParts and suppliers that change
CostRange by line item
TimelineWhen it reaches the line
Supply riskMonte Carlo on lead times
Impact cardAgent writes the explanation
Agent stepSimulationCode

Agents propose missing assumptions and flag them as estimates. Code computes every number.

05Guardrails and audit trail

Fern's agentsClaude Code sessions
Employee agentsA person's own credentials
MCP toolsEvery agent goes through these
Policy gateWho, for whom, stricter limit
Action logEvery step in SpacetimeDB
  • Web research: Claude web search first, Bright Data when a page blocks it. Code checks each proposal.
  • Audit trail: check runs are immutable and keep the seed, code version and assumptions.
  • Evidence: every number carries its class: fact, assumption, computation or judgment.

06Results

Demo scenario: ECR-0417, a cracked bracket

COST+$415Per aircraft ($142k to $156k one time)
TIME TO THE LINE13wkLater than the engineer asked
LEAD TIME14wkSupply risk: one qualified source

Cal Hacks 13.0 · Hyphenate

DELAY FOUND13wkSupplier-driven, aerospace scenario
POTENTIAL SAVINGS$180KQualification savings evaluated
ASSESSMENT TIME~60sWith auditable calculations
RESULTTop 20Out of 150 teams, Cal Hacks hackathon (Hyphenate track)
Supplier delay distributionMONTE CARLO
Distribution of supplier-driven schedule delay in weeks, from 0 to 24. The distribution peaks at the identified 13-week delay, marked on the chart.04812162024Schedule delay, weeks13 wkSupplier-driven delay
Distribution of supplier-driven schedule delay in weeks, from 0 to 24. The distribution peaks at the identified 13-week delay, marked on the chart.04812162024Schedule delay, weeks13 wkSupplier-driven delay

Distribution of supplier-driven schedule delay in weeks, from 0 to 24. The distribution peaks at the identified 13-week delay, marked on the chart.

07My role

Team project at Cal Hacks 2026, placed Top 20 out of 150 teams in the Hyphenate track. I co-developed the system that integrates PLM and ERP data and the modeling that drives the cost, supplier-risk and production-planning assessment.

Ankit Chatriki
Ankit ChatrikiCO-DEVELOPER

08Stack and links

TypeScriptClaude Agent SDKMCPSpacetimeDBBright DataFrappe / ERPNextPLMMonte Carlo
View on Devpost
© 2026 Ankit ChatrikiBuilt with Next.js, GSAP, LenisBack to top
Back to home