EarningsLens

Turn a raw earnings call transcript into clean, structured financial signals — beat/miss, guidance, and management tone — in one POST request.

EarningsLens reads an earnings call transcript (give it a URL or paste the raw text) and uses DeepSeek AI to pull out the numbers and narrative that matter: revenue and EPS actuals versus estimates, beat/miss flags, next-quarter revenue guidance, key themes, management tone, and Q&A red flags. You get a single, predictable JSON object instead of paragraphs of prose. No CSV parsing, no regex over transcript HTML, no model prompt engineering on your end.

Endpoints

POST /extract request body (JSON): supply either url (a transcript page to fetch) or text (raw transcript pasted directly) — at least one is required. Optional ticker is a hint (e.g. "NVDA") used to anchor the extraction.

Why this API

• One request, one structured object. You hand over a transcript; you get back typed JSON with the exact same field shape every time — ready to drop into a database, dashboard, or screener.
• Both numbers and narrative. Beyond revenue/EPS beat-miss math, it captures the soft signals analysts actually trade on: management_tone, guidance tone, key_themes, and qa_red_flags (analyst questions that revealed friction, deflection, or repeated pressure).
• URL or raw text. Point it at a transcript page and it fetches and strips the HTML for you, or paste raw text directly when you already have it.
• Ticker inference built in. If a transcript doesn't state the ticker, the model infers it from the company name (e.g. "NVIDIA Corporation" → NVDA), so results stay keyed even on messy sources.
• Cached for speed and cost. Identical requests are served from cache (24-hour TTL by default), so repeated lookups of the same transcript return instantly.

Typical use cases

• Earnings dashboards & screeners — populate a beat/miss table and guidance column the moment a transcript is available.
• Quant & signal pipelines — feed management_tone, guidance tone, and qa_red_flags into sentiment or event-driven models.
• Newsletters & research summaries — auto-generate a tight bullet recap (key_themes, beat/miss, next-quarter guidance) from a transcript link.
• Backfilling historical transcripts — paste archived transcript text to extract a normalized record per quarter.
• Alerting — flag calls where guidance tone is cautious or qa_red_flags is non-empty.

Good to know

• You bring the transcript. EarningsLens does not host or search a transcript library — you must supply a url to fetch or the raw text. There is no ticker/date lookup endpoint.
• AI extraction, not a filing parser. Signals are extracted by an LLM (DeepSeek) reading the transcript text. It is accurate on well-formed transcripts but is not a substitute for audited SEC filings; verify figures before relying on them for trading or reporting.
• Nullable by design. Any field the model can't determine comes back null (or an empty list for key_themes / qa_red_flags). Numbers in revenue/guidance are expressed in the stated unit (e.g. "B" for billions) and currency.
• Input is trimmed. Fetched or pasted transcripts are capped (~80,000 characters) before extraction; very long transcripts are truncated, which can drop late-call details.
• url fetching depends on the source. If the target page blocks the fetch or returns an error, /extract responds with 502; pasting the text directly avoids this.
• word_count reflects the supplied text, and source_url echoes the url you passed (or null when you posted raw text).
• Auth: access is via the RapidAPI gateway; the GET /health check is open and unauthenticated.