Dynamic Budget Suggestions with confidence scoring and trend analysis (#73) by aliraza556 · Pull Request #79 · rohitdash08/FinMind

aliraza556 · 2026-02-15T22:27:53Z

Summary

Implements the Dynamic Budget Suggestions feature (Issue #73) that suggests budget limits
using past spending patterns to improve onboarding and financial guidance.

What changed

Backend: Budget suggestion engine (services/ai.py) — Rewrote the heuristic budget algorithm to analyze 3–6 months of historical spending data using weighted averages (recent months weighted higher). Computes a confidence score based on data availability, detects per-category spending trends (increasing/decreasing/stable), and generates actionable budget limits with a 50/30/20 breakdown.
Backend: API endpoint (routes/insights.py) — Enhanced /insights/budget-suggestion with month (YYYY-MM) and months (3-6) query parameters. Added Redis caching with 30-minute TTL and input validation.
Backend: Cache invalidation (routes/expenses.py, services/cache.py) — Added budget_suggestion_key and ensured budget suggestion cache is cleared when expenses are created, updated, or deleted.
Backend: OpenAPI spec (openapi.yaml) — Updated endpoint documentation with new query parameters and response schemas (BudgetSuggestion, ConfidenceScore, CategorySuggestion).
Frontend: API client (api/insights.ts) — New TypeScript API client with full type definitions for the budget suggestion response.
Frontend: Budgets page (pages/Budgets.tsx) — Replaced static content with a dynamic dashboard showing suggested budget, spending trend, 50/30/20 breakdown, per-category suggestions with trend indicators, and monthly spending history bars. Includes 3/6 month toggle and refresh button.
Tests (tests/test_budget_suggestions.py) — 18 comprehensive tests covering no-data defaults, single/multi-month analysis, confidence scoring, trend detection (increasing/decreasing/stable), lookback parameter clamping, income exclusion, 50/30/20 breakdown accuracy, and monthly history.
Dependency (requirements.txt) — Added python-dateutil for reliable month arithmetic via relativedelta.

Closed: #73

Acceptance Criteria

Uses 3–6 month historical spending data
Provides confidence score (0-100% with labels: no_data, low, medium, high, very_high)
Per-category budget suggestions with trend analysis
Falls back to OpenAI when API key is configured
Redis caching with automatic invalidation on expense changes
18 backend tests all passing

ScreenShorts:

Test Plan

Run docker compose up --build — all services start successfully
Run docker compose run --rm backend python -m pytest -v tests/ — 28/28 tests pass
Register a new user, add expenses across 2+ months, verify budget suggestions
update with real data on the Budgets page
Toggle between 3 Months and 6 Months — verify data range updates
Click Refresh — verify fresh data is fetched
Verify confidence score increases as more months of data are added
Verify spending trends correctly show increasing/decreasing/stable


## Files Changed (9 files, +1303 / -316 lines)

| File | Change |
|------|--------|
| `app/src/api/insights.ts` | New — TypeScript API client + types |
| `app/src/pages/Budgets.tsx` | Modified — Dynamic budget dashboard |
| `packages/backend/app/openapi.yaml` | Modified — API spec with new schemas |
| `packages/backend/app/routes/expenses.py` | Modified — Cache invalidation for budget suggestions |
| `packages/backend/app/routes/insights.py` | Modified — Enhanced endpoint with params + caching |
| `packages/backend/app/services/ai.py` | Modified — Multi-month heuristic engine |
| `packages/backend/app/services/cache.py` | Modified — Added budget_suggestion_key |
| `packages/backend/requirements.txt` | Modified — Added python-dateutil |
| `packages/backend/tests/test_budget_suggestions.py` | New — 18 tests for budget suggestions |

aliraza556 · 2026-02-15T23:09:39Z

@rohitdash08 The checks have passed. Please review the code and share your feedback.

rohitdash08 · 2026-02-16T18:54:06Z

Hi @aliraza556, looks good to me but we have some modification related to your PR as per now this is not providing insights you should add persona instead of system prompt. also can you confirm did you check with your openai api key as it was failing to generate insight for the changes you made.

Could you please share your Discord?

- Replace single user prompt with system persona + user data pattern - Add response_format for reliable JSON output - Include insights field in AI response - Log OpenAI failures instead of silently swallowing errors

aliraza556 · 2026-02-16T19:58:15Z

Hi @rohitdash08, thanks for the feedback! I've fixed the OpenAI integration — added a proper FinMind persona as the system message instead of the old single-prompt approach, enabled JSON response format so it won't fail to parse, and added an insights field to the output. Also added error logging so failures are visible now instead of silently falling back. My Discord is aliraza_804, feel free to reach out there.

rohitdash08 · 2026-02-16T20:09:03Z

@aliraza556, our tests are still failing. Can you please share your test script here?

aliraza556 · 2026-02-16T20:11:41Z

Hi @rohitdash08, here's the full test script — tests/test_budget_suggestions.py. It covers all the key scenarios:

Test Categories:

Endpoint returns suggestion with no data (default budget)

Single month and multi-month data handling

Data range reflects actual months with data

Monthly totals included in response

Lookback param (3 months / 6 months / clamped values)

Custom month param accepted

Invalid month format rejected

Income excluded from suggestions

Confidence scoring (0 months = 0%, more months = higher confidence)

Trend detection (increasing / decreasing / stable)

Monthly history per category

Breakdown sums to suggested total

Default 50/30/20 breakdown ratios

How to run:
docker compose run --rm -e PYTHONPATH=/app backend pytest tests/test_budget_suggestions.py -v
All 18 tests pass locally. Can you share the error output so I can check what's failing on your end?

rohitdash08 · 2026-02-16T20:15:27Z

@aliraza556, it's only related to the budget of the consumer and all I can see here the tests are related to the changes you made.

Please resolve the comments and I still can't see openai/gemini integration here, feel free to use gemini free api here

rohitdash08 · 2026-02-16T20:20:49Z

@aliraza556, I'm having trouble to add you on discord. Can you please add me @geekster007
so that I can help you here?

aliraza556 · 2026-02-16T20:21:43Z

@aliraza556, I'm having trouble to add you on discord. Can you please add me @geekster007 so that I can help you here?

@rohitdash08 Sure

aliraza556 · 2026-02-16T20:25:00Z

@aliraza556, it's only related to the budget of the consumer and all I can see here the tests are related to the changes you made.

Please resolve the comments and I still can't see openai/gemini integration here, feel free to use gemini free api here

Hi @rohitdash08, I've added Gemini integration for the budget suggestions — it now uses the free Gemini API as the primary AI provider since we don't have an OpenAI key set. The priority chain is: Gemini -> OpenAI -> Heuristic fallback. Also added the FinMind persona, an insights field in the response, and proper error logging. All 18 budget tests + other tests still pass.

rohitdash08 · 2026-02-16T20:33:18Z

@aliraza556, again I'm saying it's not specific to the budget contraint. The priority chain is like user can provide their free gemini api key and get the insights if they don't have openai key. Persona still need to be improved we are still getting test failure for your PR

…nger valid. Google renamed it.

aliraza556 · 2026-02-16T20:49:27Z

@aliraza556, again I'm saying it's not specific to the budget contraint. The priority chain is like user can provide their free gemini api key and get the insights if they don't have openai key. Persona still need to be improved we are still getting test failure for your PR

Hi @rohitdash08, I understand — the Gemini integration is already set up as a general fallback: if user has Gemini key → use Gemini, if OpenAI key → use OpenAI, otherwise → heuristic. I've also updated the default model to gemini-2.0-flash since gemini-1.5-flash was returning 404.

rohitdash08 · 2026-02-16T21:07:11Z

@aliraza556, I think you're aligned with the requirement which I told you on discord feel free to ask please change it back to gemini-1.5-flash in your next commit

- Add 25 mocked tests for Gemini/OpenAI integration covering API calls, persona verification, fallback chain, JSON parsing, and error handling - Improve FinMind persona with structured sections (Expertise, Personality, Rules) for richer AI-generated insights - Revert default model to gemini-1.5-flash (confirmed working) - All 50 tests pass with black and flake8 clean

aliraza556 · 2026-02-16T21:10:03Z

@aliraza556, I think you're aligned with the requirement which I told you on discord feel free to ask please change it back to gemini-1.5-flash in your next commit

@rohitdash08 Done

rohitdash08 · 2026-02-16T21:20:59Z

@aliraza556, still failing the tests resolve the comments.
are you having some trouble can you created the demo for this as well?

you can use the mock data for insights testings

aliraza556 · 2026-02-20T01:50:28Z

Hi @rohitdash08, I've fixed the formatting issue and added comprehensive AI integration tests. Here's the local test run:
53/53 tests pass — includes:

18 heuristic budget tests (no data, single/multi month, trends, confidence, 50/30/20)
6 Gemini integration tests (mocked API, persona in prompt, markdown fence parsing, fallback)
3 OpenAI integration tests (mocked client, system persona, fallback)
4 fallback chain tests (Gemini → OpenAI → heuristic priority)
6 JSON parsing tests
3 persona validation tests
13 existing tests (auth, bills, categories, dashboard, expenses)

Also: black --check . and flake8 . both clean.

Model reverted to gemini-1.5-flash as requested. Could you approve the CI workflow run for the latest commit? It shows "Action required" since it's a fork PR.

rohitdash08

@aliraza556 , also please resolve the merge conflicts

rohitdash08 · 2026-02-16T20:11:16Z

+)
+
+
 def _openai_budget(uid: int, ym: str, lookback: int = MAX_MONTHS):


change the method name

any update on this?

@rohitdash08 I am on it.

rohitdash08 · 2026-02-16T20:12:08Z

    return _heuristic_budget(uid, ym, lookback)


+FINMIND_PERSONA = (


improve the persona, this should analyze the manual input as well

rohitdash08 · 2026-02-16T21:17:51Z

-    "rising you say so clearly. You always use the 50/30/20 rule "
-    "(needs/wants/savings) as a starting framework and adjust based on "
-    "the user's actual patterns. You respond ONLY with valid JSON."
+    "You are FinMind, a certified personal finance advisor and data analyst "


Don't put finmind, it should be financial analyst

rohitdash08 · 2026-02-16T21:18:17Z

+    "2. Highlight the single biggest saving opportunity first.\n"
+    "3. Provide at least 2 actionable tips tied to specific categories.\n"
+    "4. Include short motivational insight (one sentence).\n"
+    "5. Respond ONLY with valid JSON — no markdown, no commentary."


persona is still lacking @aliraza556

…tions

…nalyst persona

'- Financial analyst persona validation' in the module docstring to match the renamed class and the reviewer's branding request.

aliraza556 · 2026-03-17T20:10:23Z

Hi @rohitdash08, I've addressed all your review comments. Here's a summary of everything done:

1. Method names renamed (change the method name)

_gemini_budget → _gemini_budget_suggestion
_openai_budget → _openai_budget_suggestion

2. "FinMind" removed from persona (Don't put finmind, it should be financial analyst)

Renamed FINMIND_PERSONA → FINANCIAL_ANALYST_PERSONA
Replaced all "FinMind" branding with "certified financial analyst"

3. Persona improved with manual input handling (improve the persona, this should analyze the manual input as well)

Added a dedicated ## Manual Input Handling section so the AI incorporates user-provided context (upcoming expenses, salary changes, financial goals, life events)

4. Persona depth improved (persona is still lacking)

Expanded to 3 structured sections: Core Expertise, Personality, Rules
8 rules covering: data-driven advice, manual context blending, biggest saving opportunity first, actionable tips per category, quantifying spending increases, motivational insight, emergency-fund targeting, JSON-only output
Added expertise areas: debt-to-income ratio awareness, income vs. expense gap analysis, anomaly detection, non-judgmental coaching tone

5. Merge conflicts resolved

All 9 conflicts across 4 files resolved — HEAD's multi-month analytics merged with origin's BYOK Gemini and persona features

All 63 backend tests pass. black formatting is clean.

Please let me know if you'd like any further changes!

aliraza556 added 3 commits February 16, 2026 03:22

feat: add dynamic budget suggestions using 3-6 month spending history

f80dfd6

files reformatted

d027b38

remove unused import timedelta

51be3ac

fix: add FinMind persona and improve OpenAI insight generation

15bbf9d

- Replace single user prompt with system persona + user data pattern - Add response_format for reliable JSON output - Include insights field in AI response - Log OpenAI failures instead of silently swallowing errors

feat: add Gemini AI integration for budget suggestions

6b50f8f

The error is 404 Not Found — the model name gemini-1.5-flash is no lo…

f482599

…nger valid. Google renamed it.

Ali Raza and others added 2 commits February 20, 2026 06:14

Merge branch 'main' into feat/dynamic-budget-suggestions

b481f54

fix: apply black formatting to test_budget_suggestions.py

7c3a790

rohitdash08 suggested changes Mar 17, 2026

View reviewed changes

aliraza556 added 3 commits March 18, 2026 00:31

Merge remote-tracking branch 'origin' into feat/dynamic-budget-sugges…

0a79c5b

…tions

fix: resolve review comments — rename AI methods, improve financial a…

1174dc9

…nalyst persona

Replace '- FinMind persona validation' with

8882ff7

'- Financial analyst persona validation' in the module docstring to match the renamed class and the reviewer's branding request.

		)


		def _openai_budget(uid: int, ym: str, lookback: int = MAX_MONTHS):

		return _heuristic_budget(uid, ym, lookback)


		FINMIND_PERSONA = (

Conversation

aliraza556 commented Feb 15, 2026

Summary

What changed

Closed: #73

Acceptance Criteria

ScreenShorts:

Test Plan

Uh oh!

aliraza556 commented Feb 15, 2026

Uh oh!

rohitdash08 commented Feb 16, 2026

Uh oh!

aliraza556 commented Feb 16, 2026

Uh oh!

rohitdash08 commented Feb 16, 2026

Uh oh!

aliraza556 commented Feb 16, 2026

Uh oh!

rohitdash08 commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rohitdash08 commented Feb 16, 2026

Uh oh!

aliraza556 commented Feb 16, 2026

Uh oh!

aliraza556 commented Feb 16, 2026

Uh oh!

rohitdash08 commented Feb 16, 2026

Uh oh!

aliraza556 commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

rohitdash08 commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

aliraza556 commented Feb 16, 2026

Uh oh!

rohitdash08 commented Feb 16, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

aliraza556 commented Feb 20, 2026

Uh oh!

rohitdash08 left a comment

Choose a reason for hiding this comment

Uh oh!

rohitdash08 Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

rohitdash08 Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

aliraza556 Mar 17, 2026

Choose a reason for hiding this comment

Uh oh!

rohitdash08 Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

rohitdash08 Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

rohitdash08 Feb 16, 2026

Choose a reason for hiding this comment

Uh oh!

aliraza556 commented Mar 17, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

rohitdash08 commented Feb 16, 2026 •

edited

Loading

aliraza556 commented Feb 16, 2026 •

edited

Loading

rohitdash08 commented Feb 16, 2026 •

edited

Loading

rohitdash08 commented Feb 16, 2026 •

edited

Loading