Nvidia rolls out new AI models to cut costs and boost performance

The post South Korea Launches Innovative Stablecoin Initiative appeared on BitcoinEthereumNews.com. South Korea has witnessed a pivotal development in its cryptocurrency landscape with BDACS introducing the nation’s first won-backed stablecoin, KRW1, built on the Avalanche network. This stablecoin is anchored by won assets stored at Woori Bank in a 1:1 ratio, ensuring high security. Continue Reading:South Korea Launches Innovative Stablecoin Initiative Source: https://en.bitcoinhaber.net/south-korea-launches-innovative-stablecoin-initiative

The US pauses a $41B UK tech and AI deal as trade talks stall, with disputes over food standards, market access, and rules abroad.   The US has frozen a major tech

Turn lengthy earnings call transcripts into one-page insights using the Financial Modeling Prep APIPhoto by Bich Tran Earnings calls are packed with insights. They tell you how a company performed, what management expects in the future, and what analysts are worried about. The challenge is that these transcripts often stretch across dozens of pages, making it tough to separate the key takeaways from the noise. With the right tools, you don’t need to spend hours reading every line. By combining the Financial Modeling Prep (FMP) API with Groq’s lightning-fast LLMs, you can transform any earnings call into a concise summary in seconds. The FMP API provides reliable access to complete transcripts, while Groq handles the heavy lifting of distilling them into clear, actionable highlights. In this article, we’ll build a Python workflow that brings these two together. You’ll see how to fetch transcripts for any stock, prepare the text, and instantly generate a one-page summary. Whether you’re tracking Apple, NVIDIA, or your favorite growth stock, the process works the same — fast, accurate, and ready whenever you are. Fetching Earnings Transcripts with FMP API The first step is to pull the raw transcript data. FMP makes this simple with dedicated endpoints for earnings calls. If you want the latest transcripts across the market, you can use the stable endpoint /stable/earning-call-transcript-latest. For a specific stock, the v3 endpoint lets you request transcripts by symbol, quarter, and year using the pattern: https://financialmodelingprep.com/api/v3/earning_call_transcript/{symbol}?quarter={q}&year={y}&apikey=YOUR_API_KEY here’s how you can fetch NVIDIA’s transcript for a given quarter: import requestsAPI_KEY = "your_api_key"symbol = "NVDA"quarter = 2year = 2024url = f"https://financialmodelingprep.com/api/v3/earning_call_transcript/{symbol}?quarter={quarter}&year={year}&apikey={API_KEY}"response = requests.get(url)data = response.json()# Inspect the keysprint(data.keys())# Access transcript contentif "content" in data[0]: transcript_text = data[0]["content"] print(transcript_text[:500]) # preview first 500 characters The response typically includes details like the company symbol, quarter, year, and the full transcript text. If you aren’t sure which quarter to query, the “latest transcripts” endpoint is the quickest way to always stay up to date. Cleaning and Preparing Transcript Data Raw transcripts from the API often include long paragraphs, speaker tags, and formatting artifacts. Before sending them to an LLM, it helps to organize the text into a cleaner structure. Most transcripts follow a pattern: prepared remarks from executives first, followed by a Q&A session with analysts. Separating these sections gives better control when prompting the model. In Python, you can parse the transcript and strip out unnecessary characters. A simple way is to split by markers such as “Operator” or “Question-and-Answer.” Once separated, you can create two blocks — Prepared Remarks and Q&A — that will later be summarized independently. This ensures the model handles each section within context and avoids missing important details. Here’s a small example of how you might start preparing the data: import re# Example: using the transcript_text we fetched earliertext = transcript_text# Remove extra spaces and line breaksclean_text = re.sub(r'\s+', ' ', text).strip()# Split sections (this is a heuristic; real-world transcripts vary slightly)if "Question-and-Answer" in clean_text: prepared, qna = clean_text.split("Question-and-Answer", 1)else: prepared, qna = clean_text, ""print("Prepared Remarks Preview:\n", prepared[:500])print("\nQ&A Preview:\n", qna[:500]) With the transcript cleaned and divided, you’re ready to feed it into Groq’s LLM. Chunking may be necessary if the text is very long. A good approach is to break it into segments of a few thousand tokens, summarize each part, and then merge the summaries in a final pass. Summarizing with Groq LLM Now that the transcript is clean and split into Prepared Remarks and Q&A, we’ll use Groq to generate a crisp one-pager. The idea is simple: summarize each section separately (for focus and accuracy), then synthesize a final brief. Prompt design (concise and factual) Use a short, repeatable template that pushes for neutral, investor-ready language: You are an equity research analyst. Summarize the following earnings call sectionfor {symbol} ({quarter} {year}). Be factual and concise.Return:1) TL;DR (3–5 bullets)2) Results vs. guidance (what improved/worsened)3) Forward outlook (specific statements)4) Risks / watch-outs5) Q&A takeaways (if present)Text:<<<{section_text}>>> Python: calling Groq and getting a clean summary Groq provides an OpenAI-compatible API. Set your GROQ_API_KEY and pick a fast, high-quality model (e.g., a Llama-3.1 70B variant). We’ll write a helper to summarize any text block, then run it for both sections and merge. import osimport textwrapimport requestsGROQ_API_KEY = os.environ.get("GROQ_API_KEY") or "your_groq_api_key"GROQ_BASE_URL = "https://api.groq.com/openai/v1" # OpenAI-compatibleMODEL = "llama-3.1-70b" # choose your preferred Groq modeldef call_groq(prompt, temperature=0.2, max_tokens=1200): url = f"{GROQ_BASE_URL}/chat/completions" headers = { "Authorization": f"Bearer {GROQ_API_KEY}", "Content-Type": "application/json", } payload = { "model": MODEL, "messages": [ {"role": "system", "content": "You are a precise, neutral equity research analyst."}, {"role": "user", "content": prompt}, ], "temperature": temperature, "max_tokens": max_tokens, } r = requests.post(url, headers=headers, json=payload, timeout=60) r.raise_for_status() return r.json()["choices"][0]["message"]["content"].strip()def build_prompt(section_text, symbol, quarter, year): template = """ You are an equity research analyst. Summarize the following earnings call section for {symbol} ({quarter} {year}). Be factual and concise. Return: 1) TL;DR (3–5 bullets) 2) Results vs. guidance (what improved/worsened) 3) Forward outlook (specific statements) 4) Risks / watch-outs 5) Q&A takeaways (if present) Text: <<< {section_text} >>> """ return textwrap.dedent(template).format( symbol=symbol, quarter=quarter, year=year, section_text=section_text )def summarize_section(section_text, symbol="NVDA", quarter="Q2", year="2024"): if not section_text or section_text.strip() == "": return "(No content found for this section.)" prompt = build_prompt(section_text, symbol, quarter, year) return call_groq(prompt)# Example usage with the cleaned splits from Section 3prepared_summary = summarize_section(prepared, symbol="NVDA", quarter="Q2", year="2024")qna_summary = summarize_section(qna, symbol="NVDA", quarter="Q2", year="2024")final_one_pager = f"""# {symbol} Earnings One-Pager — {quarter} {year}## Prepared Remarks — Key Points{prepared_summary}## Q&A Highlights{qna_summary}""".strip()print(final_one_pager[:1200]) # preview Tips that keep quality high: Keep temperature low (≈0.2) for factual tone. If a section is extremely long, chunk at ~5–8k tokens, summarize each chunk with the same prompt, then ask the model to merge chunk summaries into one section summary before producing the final one-pager. If you also fetched headline numbers (EPS/revenue, guidance) earlier, prepend them to the prompt as brief context to help the model anchor on the right outcomes. Building the End-to-End Pipeline At this point, we have all the building blocks: the FMP API to fetch transcripts, a cleaning step to structure the data, and Groq LLM to generate concise summaries. The final step is to connect everything into a single workflow that can take any ticker and return a one-page earnings call summary. The flow looks like this: Input a stock ticker (for example, NVDA). Use FMP to fetch the latest transcript. Clean and split the text into Prepared Remarks and Q&A. Send each section to Groq for summarization. Merge the outputs into a neatly formatted earnings one-pager. Here’s how it comes together in Python: def summarize_earnings_call(symbol, quarter, year, api_key, groq_key): # Step 1: Fetch transcript from FMP url = f"https://financialmodelingprep.com/api/v3/earning_call_transcript/{symbol}?quarter={quarter}&year={year}&apikey={api_key}" resp = requests.get(url) resp.raise_for_status() data = resp.json() if not data or "content" not in data[0]: return f"No transcript found for {symbol} {quarter} {year}" text = data[0]["content"] # Step 2: Clean and split clean_text = re.sub(r'\s+', ' ', text).strip() if "Question-and-Answer" in clean_text: prepared, qna = clean_text.split("Question-and-Answer", 1) else: prepared, qna = clean_text, "" # Step 3: Summarize with Groq prepared_summary = summarize_section(prepared, symbol, quarter, year) qna_summary = summarize_section(qna, symbol, quarter, year) # Step 4: Merge into final one-pager return f"""# {symbol} Earnings One-Pager — {quarter} {year}## Prepared Remarks{prepared_summary}## Q&A Highlights{qna_summary}""".strip()# Example runprint(summarize_earnings_call("NVDA", 2, 2024, API_KEY, GROQ_API_KEY)) With this setup, generating a summary becomes as simple as calling one function with a ticker and date. You can run it inside a notebook, integrate it into a research workflow, or even schedule it to trigger after each new earnings release. Free Stock Market API and Financial Statements API... Conclusion Earnings calls no longer need to feel overwhelming. With the Financial Modeling Prep API, you can instantly access any company’s transcript, and with Groq LLM, you can turn that raw text into a sharp, actionable summary in seconds. This pipeline saves hours of reading and ensures you never miss the key results, guidance, or risks hidden in lengthy remarks. Whether you track tech giants like NVIDIA or smaller growth stocks, the process is the same — fast, reliable, and powered by the flexibility of FMP’s data. Summarize Any Stock’s Earnings Call in Seconds Using FMP API was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story