snapshot-0611

Snapshot (Week 15 – 2025.06.11)

1. Concurrency Handling in Prediction & Explanation APIs (#16)

• Fixed potential race conditions during concurrent API calls to /stock-info/pred and /stock-info/exp.

• Both Prediction and Explanation tables are uniquely constrained by (ticker_id, predicted_date, horizon_days), which could raise IntegrityError under concurrent inserts.

• Wrapped insertion logic in try/except blocks:

  • On IntegrityError, transaction is rolled back and a fresh query is issued to retrieve the existing cache.

• Ensures consistent behavior and avoids 500 errors even under high request concurrency.

• Improved robustness for real-time multi-user scenarios and ngrok-based external inference pipelines.

• Also normalized API URL handling to avoid trailing slash issues.

---

Related PRs

• [#16] Prevent duplicate insertions by handling race conditions in prediction and explanation

snapshot-0608

Snapshot (Week 14 - 2025.06.08)

1. XAI Explanation API Integration (#14)

• Replaced mock generate_explanation() function with actual inference.
• Integrated with the ngrok-hosted Colab server to fetch real explanation results (tokens, token_scores).
• Added full SQLAlchemy model and CRUD for the Explanation table.
• Explanation results are now stored as JSON-serialized fields (Text columns).
• Defined ExplainationData schema and connected it to /stock-info/exp API response.

2. Modularization & Schema Refactor for Regression (#15)

• Split the original /stock-info endpoint into three focused APIs:

  • /stock-info/basic: Returns chart and news data
  • /stock-info/pred: Returns numerical regression prediction
  • /stock-info/exp: Returns model explanation

• Refactored PredictionData schema from classification to regression (result: float).

• Removed StockInfoWrapper and unified response format to simple JSON payloads.

---

Related PRs

• [#14] Add explanation API and model for XAI results
• [#15] Modularize stock-info endpoints and schema update for regression

snapshot-0516

Bi-Weekly Snapshot (Week 11 - 2025.05.16)

1. Model Integration and Deployment

• Connected to the Colab-hosted capston-team-5/finma-7b-4bit-quantized model API.
• Implemented the run_prediction logic and deployed prediction endpoint.
• Verified end-to-end inference pipeline and deployed with Docker Compose.

2. Database Migration and ORM Setup

• Migrated backend database from MongoDB to MySQL (Amazon RDS).
• Defined core relational models using SQLAlchemy: Ticker, ChartData, Prediction, News.
• Updated database connection and query structure to match new relational schema.
• DB schema documentation: db/README.md

3. CRUD Refactoring and DB Caching

• Refactored backend into modular CRUD files: chart.py, news.py, prediction.py, explanation.py

• Introduced get_session() helper for unified session management.

• Added DB caching logic to reduce redundant external API calls:

  • Chart: Fetches and stores only missing date ranges.
  • News: Compares latest pubDate with DB before fetching.
  • Prediction: Caches based on (ticker, horizon, predicted_date).

4. Multi-Market & Horizon-Aware Charting

• /chart and /stock-info endpoints now support:

  • Weekly data for horizon = 7
  • Monthly data for horizon = 30

• Previously always returned 30-days daily data → Now aligns better with prediction scale.

---

Related PRs

• [#8] Add KOSPI support in /search
• [#9] News integration for US/KOSPI stocks
• [#10] API connection to Colab-hosted prediction server
• [#11] Migration to MySQL + full CRUD refactor
• [#12][#13] Hotfixes: chart interval bug, Docker import path issue

Bi-Weekly Snapshot - 04/14

Bi-Weekly Snapshot (Week 7 - 2025.04.14)

1. Completed API Documentation

   • Finalized API specifications for core endpoints including /stock-info and /search.
   • Defined supported query parameters and example responses for frontend integration.

2. Initial Backend Setup and Structure

   • Set up the backend using FastAPI with a modular router structure (routers/stock.py, routers/search.py).
   • Configured Docker (Dockerfile, docker-compose.yml) and connected to MongoDB Atlas using environment variables in .env.

3. Implemented Core API Functionality

   • /stock-info: Integrated with yfinance to fetch actual stock price data (1-month period), and included placeholders for news, prediction, and XAI interpretation.
   • /search: Built an autocomplete API using MongoDB. If a keyword is provided, performs regex search over tickers and company names. If omitted, returns a curated list of 50 popular U.S. stocks.
   • Collected and inserted ticker data (ticker + name) for 518 companies from the NASDAQ-100 and S&P 500 lists into MongoDB for real-time search functionality.
   • Returned data is now connected to the real database and live price feed.

4. CORS Configuration for Development

   • Added CORS settings (CORSMiddleware) to allow frontend requests from development ports (e.g., localhost:5173), ensuring smooth API communication.