kussssh/IPO-Analyzer

docs: add detailed file-by-file execution workflow for interview preparation

added gtag

chore: max out local embedding parallel workers to 8

fix: ignore stale OpenAI embedding model env variables to prevent HuggingFace OSError

fix: use langchain_community.embeddings for HuggingFaceEmbeddings to resolve ModuleNotFoundError

revert: switch back to local sentence-transformers model and parallel embedding execution for speed

fix: poll document DB status while waiting for preparation lock so UI doesn't appear stuck on 'Preparing DRHP source documents'

fix: reduce embedding batch size to 25 and add rate limit backoff to respect OpenAI Tier 1 40k TPM limit

fix: use raw OpenAI SDK for embeddings with explicit httpx client to resolve HF Spaces APIConnectionError

fix: replace threaded embedding with sequential batching to fix APIConnectionError on HF Spaces

fix: remove custom httpx client from OpenAIEmbeddings to resolve APIConnectionError on HF Spaces

fix: reduce embed workers to 3 and increase retries to tolerate network constraints

fix: pass custom httpx client to OpenAIEmbeddings to resolve APIConnectionError

feat: migrate to OpenAI models and optimize performance

Improve Osmo chat: financial query expansion, 15 chunks, calculation-aware prompt

Fix Osmo chat: load persisted Chroma index, fall back to sparse retriever

Fix chat endpoint missing retriever function and hardcoded URL

Fix TypeScript SVG flexShrink prop error in Sidebar

Optimize local embeddings, add WhatsApp popup and pre-warming

Switch to Google Gemini free embeddings API for speed

Add new Quiz game and telemetry diagnostic tool

Optimize analysis pipeline for speed: parallel retrieval, smaller contexts, and startup pre-warming

feat: Add Google Analytics integration

add brain loader

feat: Enhance backend data conversion & add 3D Particle Brain loader

fixes prod issue

fix: increase retrieval limits and add forced patterns to M6/M7/M8 — fixes missing data across all modules

fix: resumable PDF download with Range header — handles ChunkedEncodingError on large files

fix: make label optional in _invoke_with_retry — prevents TypeError on callers without label arg

fix: httpx RemoteProtocolError on long Prospectus runs — http2=False, retry wrapper, keepalive tuning

fix: comma-aware Lakhs conversion, EBITDA null guard, high-risk alert, product copy

fix: remove fake fallback data and fix allocation_pct in analytics

fix: replace conn.executemany() with conn.execute() — not available on psycopg3 Connection

perf: batch chunk and embedding inserts to cut DB round-trips

fix: prevent pymupdf4llm deadlock on production (HF 2-CPU container)

fix: replace broken experimentalServices with proper Next.js vercel.json

fix: add frontend-next/.vercelignore to exclude node_modules from deploy

perf: parallel PDF page parsing with ThreadPoolExecutor

perf: 3x analysis speed — connection pool, retriever cache, parallel doc prep

chore: remove SQLite DB from git, stop LFS tracking for .db/.pdf files

production issue fixes

fix: wire Supabase storage/DB, auto-create schema, fix airtable sync bug

Production issue fixes

chore: migrate to Supabase storage, remove Render config

fix: remove local cache artifacts from repo

fix: production updates and removed caching error

Prod slow issue fixed

Prod slow issue fixed

fixed prod issue

fixed prod issue