A rule-based expert system built in CLIPS that advises students on budgeting, debt management, credit health, and investment readiness. Written for COMP 474/6741 — Intelligent Systems (Concordia University, Winter 2026).
This repository is a fork of DishaPadshala/IntelligentSystems. It is a group project by Sneha Khoreja and Disha Padshala. See Team & Contributions for who built what.
The system takes a student's financial profile (income, rent, expenses, savings, debt, credit utilization) as facts and applies a rule base to produce plain-language recommendations — for example, warning that rent exceeds 30% of income, flagging high-interest debt to pay down first, or confirming the student is ready to open a TFSA.
It is a symbolic, explainable AI system: every recommendation traces back to the specific rules and thresholds that fired, so a user can see why the advice was given. There is no training data and no black box — the knowledge is written down as human-readable rules sourced from public financial guidelines (FCAC, Equifax Canada, CFPB) and financial-literacy research.
Scope: 28 facts / 44 rules (Deliverable 2), roughly 1,000 lines of CLIPS across the fact base and rule base.
The system combines three classic knowledge-based AI techniques.
The core engine is a production system: knowledge is encoded as
IF conditions THEN actions rules (defrule). CLIPS uses forward chaining —
it starts from the known facts and repeatedly fires whichever rules have their
conditions satisfied, asserting new facts, until nothing more can fire. This is
data-driven reasoning: we push forward from what we know toward conclusions,
rather than starting from a goal and working backward.
Facts (student profile) ──► Rule engine (match → fire → assert) ──► Advice
Financial advice is rarely black-and-white. Certainty factors (CFs), from the MYCIN medical expert system (Shortliffe, 1976), attach a degree of belief between 0.0 and 1.0 to a conclusion. For example, "a 6-month emergency fund implies investment readiness" carries CF = 0.90 (near-certain), while a 3-month fund carries CF = 0.70 (moderate). CFs are not probabilities — they express how strongly the evidence supports a conclusion, and they let the system say "probably" instead of forcing a hard yes/no.
Real thresholds are fuzzy: is $1,499 in savings "minimal" and $1,501 "adequate"?
Fuzzy logic (Zadeh, 1965) replaces sharp boundaries with linguistic
variables — savings ∈ {none, minimal, adequate, excellent},
credit ∈ {excellent, good, fair, poor} — defined by overlapping membership
functions. A savings amount can belong partially to two categories at once,
which mirrors how a human advisor actually reasons about "roughly enough."
Facts (the what) are deliberately separated from rules (the how to reason). This keeps domain thresholds editable in one place without touching inference logic — a standard knowledge-engineering practice.
src/
├── facts/
│ ├── financial_facts.clp # Core thresholds: budget %, debt/credit, savings targets
│ ├── certainty_facts.clp # CF weights for stress, default risk, investment readiness
│ └── fuzzy_facts.clp # Membership-function parameters + fuzzy templates
├── rules/
│ ├── financial_rules.clp # D1 crisp rules: budgeting, rent, debt, credit, TFSA
│ ├── certainty_rules.clp # CF rules: financial stress, default risk, investment readiness
│ └── fuzzy_rules.clp # Fuzzy rules: savings × credit and income × debt combinations
├── demo.clp # Loads facts + rules with a full sample student profile
└── test_scenario.clp # Standalone test student for validation
docs/
├── D1_IMPROVEMENTS.md # Deliverable-1 feedback and how D2 addressed each item
└── IMPLEMENTATION.md # Implementation progress notes
Separation of concerns:
- Facts hold tunable domain knowledge (e.g.
max-rent-percent 30,ready-cf emergency-fund-6months 0.90,fuzzy-savings-adequate-low 1000). - Rules hold the reasoning. A budgeting analyst can adjust a threshold in the fact base without ever reading the rule logic.
The crisp (D1) and certainty-factor layers run in standard CLIPS. The fuzzy
layer uses deftemplate membership functions and requires FuzzyCLIPS.
Crisp + certainty-factor demo (CLIPS):
; from the repository root
(load "src/facts/financial_facts.clp")
(load "src/rules/financial_rules.clp")
(load "src/test_scenario.clp")
(reset)
(run)Or load the bundled demo profile:
(load "src/demo.clp")
(reset)
(run)Fuzzy + certainty layers (FuzzyCLIPS):
(load "src/facts/fuzzy_facts.clp")
(load "src/facts/certainty_facts.clp")
(load "src/rules/fuzzy_rules.clp")
(load "src/rules/certainty_rules.clp")
(reset)
(run)Get CLIPS/FuzzyCLIPS: https://clipsrules.net (FuzzyCLIPS is a fuzzy-logic extension of the CLIPS shell).
A student earning $1,500/month with $700 rent, $100 savings, and $5,000 debt triggers rules such as:
WARNING: Rent is 47% of income (threshold: 30%)
PRIORITY: Pay down high-interest debt first (18% APR)
WARNING: Credit utilization 45% — risky range (30–50%)
[Fuzzy] Minimal savings — build $1000 emergency fund before investing
Every line is traceable to the rule and threshold that produced it.
This is a group project, forked from Disha Padshala's repository. Attribution below reflects the actual commit history.
| Contributor | Role |
|---|---|
| Disha Padshala (@DishaPadshala) | Repository origin; crisp financial rule base; certainty rules 1–4; fuzzy rules 1–4; income/debt fuzzy templates; integrated demo |
| Sneha Khoreja (@sneha4175) | Core fact base; certainty rules 5–7 and their CF facts; fuzzy rules 5–8 and savings/credit membership parameters; Deliverable-2 merge integration |
| Siya Patel (@Siyapatel2704) | Early edits to the financial rule file |
Sneha Khoreja's specific contributions:
src/facts/financial_facts.clp— the core fact base (budget, debt/credit, and savings thresholds).- Certainty rules 5–7 in
src/rules/certainty_rules.clp— investment-readiness (high/moderate confidence) and budget-success prediction — plus the matching CF facts insrc/facts/certainty_facts.clp. - Fuzzy rules 5–8 in
src/rules/fuzzy_rules.clp— savings × credit combinations — plus the savings and credit-health membership parameters insrc/facts/fuzzy_facts.clp. - Early crisp rules on high-interest debt and credit-utilization warnings, and resolving the Deliverable-2 merge that combined both members' work.
Deliverable 1: 117/140 (83.5%). Feedback centered on repository structure,
system size, and demo interactivity. Deliverable 2 addressed these by
reorganizing the folder hierarchy, expanding to 28 facts / 44 rules, and adding
the certainty-factor and fuzzy-logic layers. See
docs/D1_IMPROVEMENTS.md for the full feedback-to-fix
mapping.
- Shortliffe, E. H. (1976). Computer-Based Medical Consultations: MYCIN. Elsevier.
- Zadeh, L. A. (1965). Fuzzy Sets. Information and Control, 8(3), 338–353.
- Financial Consumer Agency of Canada (FCAC) — emergency-fund and budgeting guidelines.
- Equifax Canada / CFPB — credit-utilization thresholds.