From ChatGPT-4 to ChatGPT-5: The Finlee AI Upgrade at Chartledge
Published on August 10, 2025, 6:36 AM
🔈 Reading...
Abstract
This article described Chartledge’s upgrade of Finlee AI from ChatGPT-4 to ChatGPT-5 and explained how users could engage with the system to get reliable portfolio insights, visual analytics, news briefs, and professional-style recommendations. On the user side, Finlee supported global (non–user-segregated) SavedKPI data, ticker discovery, portfolio summaries, KPI charting, basic price projections, and optional buy/sell/hold guidance. The paper also outlined data integrity measures—such as timestamped KPI anchoring to a public blockchain proof system—plus privacy, limits, and best-practice prompts. Together, these changes aimed to deliver faster, clearer answers while maintaining auditability and a friendly, step-by-step experience for both guests and signed-in members.
Keywords: Finlee AI, Chartledge, GPT-5, portfolio analytics, KPIs, blockchain anchoring
Introduction
Finlee AI is Chartledge’s conversational assistant for investment literacy and portfolio analysis. In 2025, Chartledge upgraded Finlee’s core model from ChatGPT-4 to ChatGPT-5. The goal was simple: keep the product’s voice and workflow familiar while improving reasoning quality, consistency, and the breadth of supported tasks. Rather than a cosmetic change, the upgrade was paired with a careful re-platforming of Finlee’s back end to the newer OpenAI client patterns and response flow, providing stability today and flexibility for future features.
This paper provided (a) a high-level look at what changed under the hood, and (b) a practical, hands-on guide for using Finlee to get the most out of Chartledge—whether users were guests exploring the “Chartledge 30” or members tracking their own holdings.
What Changed Under the Hood
Modernized OpenAI Integration
Finlee now initialized the OpenAI client using the current “Responses API” approach for GPT-5 and automatically fell back to legacy chat completions if an older environment was detected. This ensured reliable behavior across deployments while standardizing how messages and tokens were handled.
Consistent Behavior for Everyone
SavedKPI data were treated as a global pool for answering questions. That meant users could query any ticker present in the database without worrying about per-user partitions. Guests were supported via an admin-hosted configuration, while signed-in users continued to receive a tailored experience.
Safer, Auditable Data Flow
When Finlee summarized recent KPIs for a ticker, the exact weekly batch used could be hashed and anchored via OpenTimestamps, producing a durable proof of “what data, when.” This helped auditors and power users verify that an answer matched a specific dataset at a specific time.
Using Finlee AI: A Step-by-Step Guide
1. Access and Community Support
Open the Finlee AI page on Chartledge. Finlee remained free to use because it relied on community support rather than hard request caps. Users helped sustain hosting, model costs, and new feature development by (a) signing up for a free Chartledge account and (b) posting or purchasing Chartledge NFTs. Creating an account improved personalization and reliability, while NFT activity directly funded compute and ongoing maintenance—keeping Finlee open and accessible for everyone.
2. Guests vs. Signed-In Users
Guests saw a short note that responses were based on the Chartledge 30 list and public KPIs.
Signed-in members got the same features plus continuity with their account and preferences.
3. Ask in Plain English
Finlee handled direct questions and multi-part prompts. Good starting points:
“Show my tickers.” (Returned the full list of available tickers in the SavedKPI database.)
“Portfolio summary” or “Show Chartledge 30.” (Generated a weekly performance overview.)
“AAPL chart for price, percent change.” (Plotted recent KPIs as inline charts.)
“BTC 24h high and low” or “ETH volume and price.” (Surfaced the requested KPIs.)
“What’s the latest news on the Federal Reserve?” (A brief, plain-language news summary.)
“Should I buy, sell, or hold NVDA?” (Returned a concise strategy-style recommendation.)
“Project SOL price in 2027” or “Where might ETH be in five years?” (A simple projection based on historical trends, described below.)
4. Working with KPIs
Finlee recognized a shared set of KPI fields for stocks (e.g., price, percent change, P/E, market cap, dividend yield, 52-week high/low, PEG) and crypto (e.g., price, percent change, high/low, volume). Users could ask for one or several fields in a single request. If the prompt was vague, Finlee defaulted to price and filled in sensible context.
5. Charts and Visuals
When a chart was requested (or implied), Finlee generated a clean line plot over the recent window (typically the past week). Multiple requested fields produced multiple lightweight charts. These appeared inline so users could visually check momentum, volatility, and possible inflection points.
6. Portfolio Summaries
“Portfolio summary” scanned the week’s KPIs across all available tickers, noted meaningful price changes, and returned a one-paragraph synthesis. This was especially helpful for users who wanted a fast, human-readable wrap-up without combing through charts.
7. News Briefs
If a prompt looked like general news (e.g., “latest headlines,” “what happened today,” or named political leaders and events), Finlee fetched a short list of headlines and wrote a compact, neutral summary (2–4 sentences). If the SavedKPI database was empty, Finlee defaulted to a broader news mode so users still received value.
8. Buy/Sell/Hold Guidance
If a user asked for a recommendation, Finlee combined simple momentum (recent percent change) with valuation context (such as P/E for equities when available) to deliver a Buy, Sell, or Hold signal. The output was intentionally conservative, emphasizing clarity and guardrails instead of aggressive calls. If data were thin, Finlee said so and leaned “Hold.”
9. Lightweight Projections
When the prompt included a future year (e.g., “in 2028” or “in five years”), Finlee used one of two historical methods:
CAGR-style projection over the full KPI history for the instrument, or
A simple linear trend fit to recent price points.
Results were expressed as estimates, not price targets. Finlee added context when the signal looked weak (e.g., scarce data or unstable recent behavior).
10. Text-to-Speech (Optional)
Finlee cleaned its final text (removing markdown and symbols) to provide a TTS-friendly version. This was helpful for accessibility and mobile use.
Data Integrity and Auditability
Finlee could serialize the KPI rows used for a given weekly answer, hash them, and anchor that hash using an external timestamping service. The result: an independently verifiable proof that “the answer was generated from exactly these KPIs at exactly this time.” For organizations that required lineage and reproducibility, this improved trust without exposing any sensitive credentials.
Privacy and Security
User OpenAI keys (when provided) were stored encrypted and decrypted at runtime only when needed. For guests, Finlee relied on a secure admin configuration. The system’s news and KPI logic were designed to fail safe—if a third-party call timed out, Finlee returned a helpful message rather than stalling.
Known Limits
Time scoping: Finlee prioritized current-year context for general questions to keep results timely and reduce confusion with outdated events.
Session limits: Very long sessions might hit a question cap; Finlee clearly indicated when that happened.
Data availability: If the SavedKPI database lacked depth for a ticker, charts and projections could be limited. Finlee stated this explicitly.
Best-Practice Prompts (Copy/Paste)
“Show my tickers.”
“Portfolio summary for the past week.”
“AAPL chart for price, percent change, PEG.”
“BTC price, 24h high, 24h low.”
“Summarize today’s market news in two sentences.”
“Should I buy, sell, or hold AMD this week?”
“Project ETH price in 2030 and explain the method.”
Conclusion
The migration from ChatGPT-4 to ChatGPT-5 modernized Finlee’s core engine while keeping the user experience familiar: ask a clear question, get a clear answer—often with visuals and a short narrative you could act on. Behind the scenes, standardized client patterns, graceful fallbacks, and audit-friendly data anchoring aimed to make Finlee both more dependable and more transparent. For users at every level—from curious guests to active traders—the new Finlee offered a faster path from question to insight, with the right guardrails in place.
