Murder at the
Tech Mansion

Build an AI detective with the Reasoning Layer SDK

7 steps · 4 hours · Sorts → Terms → Inference → Negation → Fuzzy → Causal → Agents

Arrow keys to navigate · Esc for overview · F for fullscreen

Getting Started

1. Create an account & get your API key

2. Clone the workshop repo & install

git clone https://gitlab.com/kortexya-pub/workshop.git
cd workshop
npm install

3. Configure your credentials in .env

REASONING_LAYER_URL=https://platform.ovh.reasoninglayer.ai/api/v1
REASONING_LAYER_TENANT_ID=your-tenant-id
REASONING_LAYER_API_KEY=your-api-key

Verify It Works

npx tsx step-01-sorts.ts

The Case

Last night, billionaire tech CEO Marcus Chen was found dead
in the study of his mansion during a private dinner party.

Cause of death: cyanide poisoning in his wine glass.

6 guests were present. No one has left the property.

Your mission: build an AI detective to find the killer.

The Suspects

The Evidence

• Poison vial found in study (wiped clean — no fingerprints)
• Victim's wine glass contains cyanide traces
• Muddy footprints from garden to study
• Hair strand on victim's collar: female DNA, 94% match confidence
• Staff member saw "dark-clothed figure near study around 8:40pm"
• James and Elena give contradicting alibis
• Study door was locked — only key holders could enter

What is the Reasoning Layer?

Key idea: data, logic, and types are all the same thing — Psi-terms.

Key Analogies

STEP 1

Setting Up the Case
Sorts

Define the types of things in your investigation.

Sorts are like classes, but with multiple inheritance that actually works.

File: step-01-sorts.ts · 25 min

STEP 1

The Sort Hierarchy

No diamond problem — the system computes GLB automatically.

GLB(suspect, witness) = person_of_interest

STEP 1

What You Build

• person, suspect, witness, person_of_interest
• evidence, physical_evidence, forensic_evidence, testimony
• location, room

Run: npx tsx step-01-sorts.ts

STEP 2

Building the Case File
Terms

Enter all 6 suspects, evidence, and testimonies as terms.

File: step-02-terms.ts · 20 min

STEP 2

Tagged Values (Storage Format)

Value.string("Alice")    →  {"type": "String",  "value": "Alice"}
Value.integer(35)        →  {"type": "Integer", "value": 35}
Value.boolean(true)      →  {"type": "Boolean", "value": true}
Value.uninstantiated()   →  {"type": "Uninstantiated"}

STEP 2

What You Build

• 6 suspect terms with name, motive, alibi, key access
• Physical evidence: poison vial, wine glass, footprints
• Forensic evidence: hair strand with DNA match
• Testimonies from each suspect
• Room locations with camera info

Can you spot the contradictions in the testimonies?

Run: npx tsx step-02-terms.ts

☕ Break — 10 min

We've covered knowledge representation (types + data).
Next: reasoning (rules + inference). That's where the magic happens.

STEP 3

The Deduction
Inference

Write logical rules. The engine deduces who could have done it.

File: step-03-inference.ts · 40 min

STEP 3

TERMS (storage):      Value.string("Alice")         → {"type":"String","value":"Alice"}
INFERENCE (reasoning): psi('suspect', {name: "Alice"}) → "Alice"

NEVER mix them.

STEP 3

The Rules

1. alibi_contradiction — A says with B at X, but B says Y
2. confirmed_alibi — Two people corroborate each other
3. had_opportunity — Has key to the study
4. matches_dna — Female + had opportunity
5. prime_suspect — Matches DNA + has motive

The engine chains them: prime_suspect → matches_dna → had_opportunity → facts

STEP 3

Backward Chaining

The engine works backward from your question — like a detective following leads.

STEP 3

Expected Results

Run: npx tsx step-03-inference.ts

CHALLENGE 3

Your Turn: Poison Access

Write a rule had_access_to_poison that identifies suspects who:

• Had access to the kitchen (alibi_location = "kitchen")
• OR had a chemistry/science background (occupation contains "Scientist")

Then query: who had access to poison AND matches DNA?

STEP 4

Eliminating the Innocent
Negation

A detective doesn't just find the guilty —
they eliminate the innocent.

NAF (Negation as Failure): if I can't prove it, it's false.

File: step-04-negation.ts · 15 min

STEP 4

Closed-World Assumption

// "Find suspects who are NOT prime suspects"
nafProve({
  literals: [
    { term: psi('suspect', {name: '?Name'}), negated: false },
    { term: psi('prime_suspect', {name: '?Name'}), negated: true },
  ]
})

Run: npx tsx step-04-negation.ts

CHALLENGE 4

Your Turn: Unverifiable Alibis

Write a NAF query to find suspects whose alibi
has no corroborating claim from anyone else.

STEP 5

Weighing Uncertain Evidence
Fuzzy Logic

Not all evidence is equally reliable.
DNA at 94% ≠ an eyewitness who "maybe saw someone."

File: step-05-fuzzy.ts · 25 min

STEP 5

Membership Functions

STEP 5

What You Build

• Value.fuzzyNumber(FuzzyShape.triangular(a, b, c))
• fuzzyUnify() — similarity degree (0.0 – 1.0)
• searchTopK() — find most similar evidence
• fuzzyProve() — inference with truth degrees

Run: npx tsx step-05-fuzzy.ts

CHALLENGE 5

Your Turn: Alibi Contradiction Evidence

Create a weighted_evidence term for the contradicting alibis:

• Description: "James and Elena's stories don't match"
• Points to both of them
• Reliability: Gaussian(0.7, 0.15)

Then compare its similarity with the DNA evidence
using fuzzyUnify. Are these evidence types related?

File: challenges/challenge-05-alibi-contradiction-evidence.ts

☕ Break — 10 min

Types, data, inference, elimination, uncertainty — done.
Next: causal reasoning — why did it happen?

STEP 6

The Motive Chain
Causal Reasoning

WHY did it happen? Reconstruct the chain of events.

File: step-06-causal.ts · 20 min

STEP 6

Pearl's Ladder of Causation

Each level is strictly more powerful than the last.

STEP 6

The Causal Graph

Run: npx tsx step-06-causal.ts

CHALLENGE 6

Your Turn: Extended Causality

1. Add a new causal chain:
   secret_will_change → financial_fear → murder_motive
   (Tom Reeves' potential motive)
2. Run rootCause again — does the picture change?
3. Counterfactual: "Would the murder have happened if
   access_to_poison AND unconfirmed_alibi were both false?"

File: challenges/challenge-06-extended-causality.ts

STEP 7

Deploy Your AI Detective
Cognitive Agents

Your detective becomes an autonomous agent
with beliefs, goals, and intentions.

File: step-07-agents.ts · 20 min

STEP 7

BDI Architecture

STEP 7

What You Build

• Holmes & Watson — two detectives with different theories
• Beliefs from engine — prime suspects, contradictions, forensics
• Investigation actions — verify alibi, check camera, cross-reference
• Agent messaging — Holmes shares evidence with Watson
• Human feedback — instructor evaluates the detective

Run: npx tsx step-07-agents.ts

The Reveal

Who killed Marcus Chen?

The evidence points to Nina Torres

• Bitter breakup motive
• Key to the study
• No alibi — 45-minute gap, no witnesses
• DNA match on victim's collar
• Security background → explains wiped fingerprints

But can you prove it beyond reasonable doubt?
The kitchen camera footage might change everything...

What You Learned

Thank you!

Resources to go further:

• Concept pages for deeper understanding
• Advanced tutorials: Clinical Trial, Loan Approval, Incident Response
• API Reference for method-level docs

Key takeaway: everything is a Psi-term.
Data = logic = constraints. That's homoiconicity.