AI Agents I Built (3/7) — Stock Director: From Discord Signals to Portfolio Scoring

Full architecture of Jipyogyeon Discord channel auto-collection, signal parsing, and portfolio-based alert classification.

Summary

  • Built an automated pipeline to collect quant trading signals from Jipyogyeon Discord channels
  • Parses Discord embed messages into structured data, matches against portfolio holdings, and classifies alert priority
  • Integrates with the home server dashboard to surface critical/high alerts and deliver daily briefings as summary cards

Background

Subscribing to quant signal services generates a flood of data, but there is no layer that quickly filters "signals relevant to my portfolio." Manually checking signals posted to Discord channels does not scale.

Jipyogyeon distributes dozens of signals per day across multiple channels — entries, exits, PEG, momentum, Top5 reports. The requirement was to separate signals that match portfolio holdings into immediate alerts, while queuing the rest for a post-market summary.

Stock Director automates this collection → classification → delivery pipeline.

Architecture

1. Overall Pipeline

Discord Channels (Jipyogyeon)
    ↓ 60s polling
discord_listener.py → signal_parser.py
    ↓ structured data
Portfolio Matching → Alert Classification
    ↓
    ├─ Immediate Alert → Discord Webhook + Home Server /api/webhooks
    └─ Log Only       → saved to daily_raw.json
    ↓ (15:50 KST)
Daily Briefing → Home Server /api/webhooks

2. discord_listener.py — Channel Polling

Uses the Discord HTTP API directly — not a library like discord.py. Messages are fetched via HTTP GET requests.

Core design constraint: read-only. All write operations — sending messages, reactions, typing indicators — are explicitly excluded. This is an intentional restriction to prevent Discord ToS violations.

Polling behavior: - Full channel sweep every 60 seconds - 1–1.5 second random delay between channels (rate limit avoidance) - Last-seen message ID tracked per channel; only new messages are processed - Only today's messages (KST) are processed — prior-day messages are skipped

Trading hours filter: Active only on Korean market hours, 09:00–16:00 KST, weekdays only. Outside these hours the listener sleeps until the next trading session starts. Message state resets at the start of each new trading day.

Channel configuration: Managed via channels.json. Currently monitors 10 channels:

Channel Content
์˜ค๋Š˜์˜์‹œ๊ทธ๋„ Daily market summary
์…‹์—…ํ˜•์„ฑ VCP / setup detection
์ง„์ž… Standard / breakout / aggressive entry signals
๋ถ„ํ• ๋งค์ˆ˜ Pyramiding / scale-in signals
์ฒญ์‚ฐ Final / breakout exit signals
๋ถ„ํ• ๋งค๋„ Partial profit-taking / scale-out
๋ชจ๋ฉ˜ํ…€ Momentum BUY / SELL
peg PEG-based signals
์ข…ํ•ฉ์•Œ๋ฆผ_4h 4-hour composite alerts
top5 Top 5 leader stock report

3. signal_parser.py — Discord Embed Parsing

Jipyogyeon signals arrive as Discord Rich Embeds, not plain text.

Embed structure:

embed.title:       conviction + signal type (e.g. "๐ŸŸก B ๐Ÿ’ฐ Standard Entry")
embed.description: ticker + price / SL / RR
embed.fields:      status (score, EMA alignment), signal tags, AI evaluation, description, metadata (market env, exchange)

The parser converts this structure into a Signal dataclass:

Field Source Example
conviction title emoji ๐ŸŸฃ→S, ๐ŸŸข→A, ๐ŸŸก→B, ๐ŸŸ →C, ๐Ÿ”ด→D
signal_type title signal name "Standard Entry", "Final Exit"
action signal_type mapping BUY / SELL / CHECK
ticker description parse "090430"
price / stop_loss / risk_reward description parse 142600 / 122851 / 2.1
score fields["์ƒํƒœ"] 60
ema_alignment fields["์ƒํƒœ"] aligned / inverted / crossed
ai_summary fields["AI ํ‰๊ฐ€"] "Honeypot pullback, 21 EMA support"
exchange fields["์ •๋ณด"] "KRX" / "NASDAQ"

Top5 reports are handled by a dedicated parser (parse_top5_embed), extracting rank, conviction, ticker, sector, signal type, price, stop-loss, and target price.

4. Portfolio-Based Alert Classification

Parsed signals are matched against holdings and watchlist tickers in portfolio.json to determine alert priority.

Alert rules:

Priority Condition Alert
critical SELL signal on a held position Immediate alert (exit warning)
high Any signal on a held position Immediate alert (portfolio hit)
high S/A conviction + score ≥ 60 + BUY Immediate alert (high-conviction entry)
low All other signals Log only (no alert)

Only signals requiring immediate action are forwarded to the Discord webhook and home server. The rest are stored in daily_raw.json for use in the daily briefing.

Investment strategy reflected: Korean equities support short / medium / long-term positions. US equities support long-term only. Crypto is not handled.

5. Home Server Integration

Signal data is delivered to the home server via webhook API.

POST /api/webhooks
{
  "content": "{signal JSON}",
  "tags": ["stock", "signal", "buy", "kr", "alert"],
  "source": "stock-director"
}

Tags are auto-generated from signal attributes: market (kr/us), action (buy/sell/check), urgency (alert/urgent), conviction (high-conviction).

Daily briefing: At 15:50 KST — just before market close — all signals collected during the session are aggregated into a summary:

{
  "date": "2026-04-06",
  "total": 42,
  "buy_count": 18,
  "sell_count": 8,
  "check_count": 16,
  "portfolio_hits": [...],
  "high_conviction_entries": [...]
}

The home server dashboard renders this briefing as a summary card. Critical/high alerts are visually highlighted.

6. Always-On Execution

discord_listener.py runs as a persistent service managed by macOS launchd. It starts automatically on system boot and restarts on crash.

External drive TCC restriction: macOS applies TCC (Transparency, Consent, and Control) permission constraints to files on external drives. Running scripts directly from an external drive via launchd can trigger permission failures. The workaround: maintain an execution copy on internal storage, synchronized with the canonical source on the external drive.

Engineering Challenges

Discord embed parsing complexity: Jipyogyeon does not use a single message format. Standard signals, PEG signals, and Top5 reports each have distinct embed structures. A single unified parser failed. Splitting into per-type parsers resolved the issue.

Rate limit management: Sweeping 10 channels in rapid succession hits Discord API rate limits. Adding a 1–1.5 second random inter-channel delay resolved this. On 429 responses, the listener waits for the duration specified in retry_after.

Korean ticker name mapping: Jipyogyeon uses English company names (e.g. "SK Hynix Inc."). Portfolio alerts are more readable with Korean names (e.g. "SKํ•˜์ด๋‹‰์Šค"). Korean names were added to portfolio.json and referenced during alert generation.

Off-hours resource waste: The initial implementation polled 24/7. No signals are published outside trading hours, making those API calls unnecessary. Adding the trading hours filter restricted execution to weekdays 09:00–16:00 KST.

Conclusion

The core function of Stock Director is separating signal from noise. Dozens of signals arrive daily; only a few require immediate attention.

The structure is straightforward: collect (listener) → parse (parser) → classify → deliver (webhook). Each stage has a single responsibility.

In investment information management, volume is not the metric that matters. What matters is how quickly and accurately signals that affect your portfolio are identified. This pipeline automates that filtering layer.

Series overview: Series index

๋Œ“๊ธ€

๋Œ“๊ธ€ ์“ฐ๊ธฐ

์ด ๋ธ”๋กœ๊ทธ์˜ ์ธ๊ธฐ ๊ฒŒ์‹œ๋ฌผ

Agent Memory Engine (2/10) — Building an AI Agent Memory System with SQLite Alone

← 1/10 SQLite Memory Engine f… ๐Ÿ“š Series Index 3/10 memcore Completed → A memory engine that runs without daemons, without dependencies, anywhere ํ•ต์‹ฌ ์š”์•ฝ This post covers how to design an AI agent memory engine (memcore) on a single-file SQLite backend. - Structure: 25 modules, ~6,464 lines, 22 DB tables - Principles: no daemon, single-file portability, no required external dependencies, host-agnostic - Core techniques: 4-layer hybrid search (topic · vector · FTS5 · LIKE), U-tag dialectic-based confidence evolution, ingestion-layer bias rejection rules What This Post Covers One approach to building AI agent memory without a server or cloud vector DB. Specifically: (1) the limitations of text-file-based memory, (2) vector DB alternatives compared and selection criteria, (3) how to layer search tiers, (4) how to evolve confidence scores over time, and (5) how to block biased memory accumulation at the ingestion stage. Design Principles: Why Single-File SQLite Text-file-based ...
Read more »

"ML Foundations (9/9) — PyTorch vs TensorFlow, and the Road to Local LLMs"

์ด๋ฏธ์ง€
← 8/9 Deep Learning Architec… ๐Ÿ“š Series Index (series end) The final part. So far we've focused on models. Now we focus on the tools that actually run them . We'll lay out the philosophical difference between PyTorch and TensorFlow, and trace how Hugging Face transformers , llama.cpp , MLX , and Ollama built the bridge to running large language models on your own machine. By the end you should have the full mental model of "download a pretrained LLM and serve it locally." 0. Learning Objectives Explain the eager-vs-graph difference between PyTorch and TensorFlow. Explain, in graph terms, how autograd automates the backward pass. Use Hugging Face transformers ' AutoModel , AutoTokenizer , and pipeline abstractions. Describe llama.cpp's GGUF quantization, INT4 inference, and CPU-first flow. Describe Apple MLX's use of unified memory and how it differs from PyTorch. Run a local LLM with Ollama and call it through an OpenAI-compatible API. ...
Read more »

"RAG Core Study (14/26) — Evaluation Sets with RAGAS & DeepEval"

์ด๋ฏธ์ง€
Series overview: Series index ๐Ÿ“š Series Index Without an evaluation set, every RAG improvement is a story, not evidence. RAG systems fail in more than one place: retrieval, context selection, answer generation, and grounding. That means "looks better" is not a metric. Part 14 explains how to build a golden dataset , how RAGAS and DeepEval fit into the workflow, and why teams must separate retrieval quality from answer quality if they want tuning decisions to hold up. 0. Prerequisites Part 13 reranking — system quality now depends on multi-stage ranking. Part 1 grounding — a correct-looking answer is not enough. Part 12 Hybrid — multiple retrieval settings need comparison on fixed data. 1. Learning Objectives Build a small but useful golden dataset for RAG. Distinguish retrieval metrics from answer metrics . Use RAGAS and DeepEval in the right roles. Avoid the common traps in judge-model-based evaluation. 2. ํ•ต์‹ฌ ์š”์•ฝ An evaluation set for RAG is ...
Read more »

"ML Foundations (8/9) — Deep Learning Architectures: CNN, RNN, Attention"

์ด๋ฏธ์ง€
← 7/9 Deep Learning Training ๐Ÿ“š Series Index 9/9 PyTorch vs TensorFlow → What did we keep adding on top of an MLP? Vision went CNN, sequences went RNN/LSTM, and eventually both converged on Attention and the Transformer. Each architecture is easier to remember if you read it as what it gave up to gain something else . This part is the one-line summary of the decisive inflection points: LeNet → AlexNet → ResNet → LSTM → Attention → Transformer. 0. Learning Objectives Explain why CNN's convolution, pooling, and weight sharing match images so well. Trace what got unblocked from LeNet → AlexNet → VGG → ResNet as depth grew. State BPTT for RNNs and the vanishing/exploding gradient problem. Write the LSTM gate equations (forget, input, output) with intuition. Write the attention formula in query/key/value form and the scaled dot-product variant. State the precise reasons Transformers replaced RNNs (parallelism + long-range dependencies). 1. ํ•ต์‹ฌ ์š”์•ฝ CNN : weight sharin...
Read more »

"ML Foundations (7/9) — Deep Learning Training: Optimizers, Regularization, Initialization"

← 6/9 Neural Networks ๐Ÿ“š Series Index 8/9 Deep Learning Architec… → Building an MLP, as we did in Part 6, is not the same as getting it to train well . The right optimizer, regularization, initialization, and learning rate — when these line up, deep networks converge. When they don't, the network refuses to learn at all. This part is the catalog of those crafts, with formulas, paper citations, and code in one place. 0. Learning Objectives Compare and write the update rules for SGD, Momentum, Nesterov, and Adam. Explain how Dropout, BatchNorm, and LayerNorm work and where they belong in a model. Derive the variance formulas for Xavier and He initialization and match them to activations. Implement step, cosine, and warmup learning-rate schedules in PyTorch. Explain why gradient clipping is effectively required for RNNs and Transformers. Diagnose the most common training failures (NaN, plateau, overfitting) and apply first-line fixes. 1. ํ•ต์‹ฌ ์š”์•ฝ SGD : \(w \leftarro...
Read more »

OpenClaw to Hermes Migration (2/13) — What to Preserve, Partially Port, or Discard

← 1/13 Current Structure Inve… ๐Ÿ“š Series Index 3/13 What Moves to Hermes → A code review record classifying operational scripts into three tiers: preserve, partial-preserve, and discard. What This Post Covers An asset classification framework (preserve / partial-preserve / discard) for migrating a legacy agent system to a successor A concrete case applying the Strangler Fig + Subset Migration strategy to a memory pipeline Classification results evaluated across LOC, external dependencies, and replaceability axes A method for slimming down a bloated single file (1,310 LOC) using the migration as a natural entry point Problem Definition: Port Everything or Cut Deep? The OpenClaw inventory comprises multiple agents, multiple scheduler/daemon processes, and multiple operational scripts. When moving to the successor system, Hermes, the choice is not binary. A full port carries accumulated technical debt along with it; a minimal port discards hard-won domain logic. Code-revie...
Read more »

AI Agents I Built (5/7) — Building an Automated Blogger API Publishing System

← 4/7 My Life Organizer Agent ๐Ÿ“š Series Index 6/7 Dual → Designing a Markdown-Based Auto-Publish Pipeline with Blogger API v3 + OAuth 2.0 Key Summary Blogger API v3 + OAuth 2.0 enables a pipeline that converts Markdown files into publishable HTML posts blogger_publish.py handles the full flow — frontmatter parsing → body cleanup → HTML conversion → CSS injection → API post — in a single script For bulk publishing, quota management (delay, retry, TOC suppression) is the critical design concern Platform Selection Criteria API access is mandatory for AI agent-driven content publishing. A comparison of platforms: Platform API Cost AdSense Decision WordPress.com REST API Paid plan required Paid only Cost overhead Ghost REST API Self-hosted or paid Manual setup Ops overhead Blogger v3 API Free Native integration Adopted Blogger is free, exposes a REST API, and integrates natively with AdSense. For personal blog automation, this combination i...
Read more »