Spaces:
Running
Running
Commit History
-
Fixed some minor issues where if the csv was empty then the website crashed
Pulastya B8a420cb
-
Minor Bug fixes where the NaN values in the Dataset caused issues in the Schemas used by the Agent
Pulastya Bab2c6db -
Fixed an issue due to which the frontend never recieved that the event was completed and was stuck in an infinite loop
Pulastya B2e3162d
-
Fixed File path not found issues and Multi-user issues
Pulastya B5ce70d3
-
Fixed Bugs where the SSE Streaming was improper added support for Auto Gluon, Fixed instances where stale schemas was causing EDA plots generation to fail
Pulastya B1ca2e0e
-
Fixed all output path issues
Pulastya B2cf9e11
-
Align HuggingFace API usage with official docs pattern
Pulastya B9c0055f -
Add better error logging to HF export endpoint
Pulastya B1412b09 -
Add better error logging to HF export endpoint
Pulastya Bccd6d0c -
Fix HuggingFace export: correct method signatures and add tmp directory
Pulastya B3e5a750 -
Fix HuggingFace export: correct class name and constructor
Pulastya B964d3d6 -
Fixed the Huggingface Hub Error
Pulastya B39b6341 -
Directly making API calls to the server -V3
Pulastya B3075bc4 -
Directly making API calls to the server -V2
Pulastya Bd75ca00
-
Fixing the issues with the SSE streaming all the history
Pulastya B1c52933
-
Added the functionality for users to save their generated assets in huggingface spaces using write tokens
Pulastya B2a82ed5 -
Add HuggingFace storage integration - users can now persist datasets, models, and plots to their own HuggingFace account
Pulastya B7af9e82
-
Fix infinite thinking when no file uploaded - properly handle analysis_failed SSE event
Pulastya Ba0e15f1 -
Fix Supabase auth on HuggingFace with runtime config injection
Pulastya B8154116 -
CRITICAL: Fixed race conditions, session cleanup, SSE leaks, and added localStorage persistence
Pulastya B- Bug #1: Added thread-local storage + SessionState wrapper to prevent race conditions while keeping heavy components shared - Bug #3: Implemented TTL-based session cleanup (60min timeout) with LRU eviction - Bug #4: Added localStorage persistence for chat history - survives page refresh - Bug #5: Fixed SSE connection leaks with proper cleanup flags and 50ms delay - Bug #6: Sessions now expire after 60 minutes of inactivity - Increased session cache from 10 to 50 for better scalability - Added request count tracking per session for monitoring2f3df85 -
Fixed multi-session support and follow-up queries - Session UUID reuse, multi-chat isolation, proper SSE switching
Pulastya B187c5e0 -
Fixed slow Agent loading which caused the Agent took too long to respond
Pulastya Be29cf28 -
Fixed Scalability issues
Pulastya B08646ab
-
Fixed known model training issues
Pulastya B04d51fb -
Fixed the File Path Issues
Pulastya B94bbef1
-
feat: Add 4 major system improvements - semantic layer, error recovery, token budget, parallel execution
Pulastya B05a3c74
-
Fix JSON serialization error for numpy types in SSE stream
Pulastya Be93629e
-
Real-time streaming: /run-async with BackgroundTasks, UUID-first, SSE-delivered results
Pulastya B565097f -
Fix SSE double-emit: eliminate numeric session identity, use UUID only
Pulastya Bc14102f -
ARCHITECTURAL FIX: Use single UUID session identity everywhere - removed http_session_key hack
Pulastya B29c9177 -
FIX SSE FINALLY: Use queue.empty() check instead of catching non-existent asyncio.QueueEmpty
Pulastya Be2e1cea -
FIX SSE BLOCKING: Use get_nowait() polling instead of blocking get() to avoid event loop deadlock
Pulastya B132976b -
DEBUG: Log before queue.get() to see if loop executes
Pulastya B5af048e -
FIX SSE: Register queue BEFORE yielding connection event to capture all events
Pulastya B06f2512 -
FIX SSE TIMING: Subscribe BEFORE sending connection event to capture all events
Pulastya B76ac02c -
ADD: Logging to trace SSE event streaming to client
Pulastya B408337e -
FIX CRITICAL: Remove emoji from print statements causing Unicode encoding errors
Pulastya B6222009 -
ADD COMPREHENSIVE DEBUG LOGGING - will show exactly where SSE breaks
Pulastya B1750ce2 -
CRITICAL FIX: Sync session IDs between HTTP and orchestrator for SSE
Pulastya BRoot cause: Session ID mismatch! - Frontend connects to SSE with session_id='1' (from HTTP request) - Orchestrator emits events to session.session_id='uuid-...' (internal UUID) - Events went to wrong stream, frontend received nothing Solution: - Added http_session_key property to orchestrator - app.py sets agent.http_session_key = session_key before analyze() - All emit() calls now use: getattr(self, 'http_session_key', None) or self.session.session_id - Events now emit to correct stream that frontend is listening to This ensures orchestrator emits to the SAME session_id that frontend connects to3a3eba7 -
CRITICAL FIX: Implement real SSE event streaming with ProgressManager
Pulastya B- Created src/progress_manager.py with global event broadcaster - Updated progress_callback to emit events via progress_manager - Fixed SSE endpoint to stream events from progress_manager.subscribe() - Updated frontend to handle onmessage events with proper debugging - Added comprehensive logging: connection, received events, cleanup - Events now properly flow: backend print -> emit -> SSE stream -> frontend109a48b -
Replace polling with Server-Sent Events (SSE) for real-time progress updates
Pulastya B711046d
-
Fix SyntaxError: use attribute assignment instead of global agent recreation
Pulastya B4eacfaa -
Fix model metrics display, add baseline comparison, improve formatting & progress indicators
Pulastya Bf5a1bc3 -
Fix session memory: preserve uploaded files for follow-up requests
Pulastya Bb213e33 -
Add Mistral AI integration - 1B tokens/month (33x Groq limits)
Pulastya BNEW PROVIDER: Mistral AI - Token limits: 1 billion tokens/month (~33M tokens/day) - 500K tokens/minute (42x better than Groq's 12K TPM) - 500 requests/minute (17x better than Groq's 30 RPM) - OpenAI-compatible API (minimal code changes) CHANGES: 1. requirements.txt: Added mistralai>=1.0.0 2. .env: - LLM_PROVIDER=mistral (new default) - MISTRAL_API_KEY env var - MISTRAL_MODEL=mistral-large-latest 3. orchestrator.py: - Added Mistral client initialization - Set TPM=500K, RPM=500 limits - Added mistral provider to API call logic - Reuses Groq code path (OpenAI format) - Token budget management applies to both 4. api/app.py: - Updated startup to default to mistral - Compact prompts enabled for mistral/groq BENEFITS: - 33x more daily tokens than Groq (no TPD exhaustion) - 42x higher TPM limit (no minute-level throttling) - Production-ready for heavy usage - Backward compatible (groq/gemini still work) ARCHITECTURE IMPACT: Minimal (~80 lines) - Mistral uses…be4eb33
-
fix: Support follow-up requests without re-uploading file
Pulastya B- Made file parameter optional in /run endpoint - Follow-up requests use session memory to resolve file path - Fixes 422 error when asking follow-up questions - Enables: Upload Report Train Model workflowba50130 -
feat: Add dynamic prompt system for small context window models (Groq support)
Pulastya B- Created dynamic_prompts.py with intent-based tool loading - Reduces prompt from ~20K to ~2K tokens (90% reduction) - Auto-detects user intent and loads only relevant tools - Enables Groq support without context overflow - Automatically enabled for Groq, optional for Gemini - Supports: viz-only, profiling, ML training, code execution - Fixed 'output' parameter hallucination with general correctionb8bcf55 -
fix: Support both GOOGLE_API_KEY and GEMINI_API_KEY env vars for HuggingFace compatibility
Pulastya Bfde5dd3 -
fix: Remove system_instruction parameter from chat endpoint
Pulastya B- system_instruction not supported in google-generativeai SDK - Prepend system message to first user message instead - Add safety settings to chat endpoint - Fixes: TypeError unexpected keyword argument 'system_instruction' - Chat endpoint now works with current Gemini SDKeeeb8b7
-
fix: Fix module import paths for Render deployment
Pulastya B- Change all absolute imports to relative imports within src package - Fix orchestrator.py: cache.cache_manager -> .cache.cache_manager - Fix tools/__init__.py and auto_pipeline.py imports - Fix all utils imports in tools/ directory (35 files) - Fix api/app.py to use src.orchestrator import - Resolves ModuleNotFoundError on Render deployment227cb22