2025 02 27 dhwani research milestones
Dhwani - Research Milestone
3 Months Plan
Key Activities
- Scaling and Verifying Concurrent Users: Ensure the system can handle multiple users simultaneously.
- Rate Limiting: Implement measures to control the rate of requests to prevent system overload.
- Multi-Language Support - Batching: Enable support for multiple languages and optimize processing through batching.
- Immersive Voice Mode: Develop a mode for teaching, entertainment, and exploration with system prompts.
- Fine-Tuning Models: Continuously improve the models based on feedback and performance data.
- Automated Red Teaming: Simulate attacks to test and improve the system's security.
- Weekly Progress Updates: Provide updates on techniques tried, comparisons against top providers, and cost metrics.
Month 1
Week 1
- API Standards: Define and implement API standards for the project.
- Logging and Automatic Configuration of GPU: Set up logging and automatic configuration for GPU resources.
Week 2
- Performance Measurement: Measure the performance of the models.
- Eval Benchmarks: Establish benchmarks for evaluation and comparison.
Week 3
- Encryption and Privacy Management: Implement encryption and privacy management protocols.
Week 4
- Delta Updates to Models: Apply delta updates to the models for continuous improvement.
- RLHF and Federated Learning: Implement Reinforcement Learning from Human Feedback (RLHF) and federated learning techniques.
- Open Data Collection: Collect open data for training and validation.
- Weekly Cost Metrics Export: Export and analyze weekly cost metrics.
- Newsletter Enrollment: Enroll users in a newsletter for regular updates and engagement.
Month 2
Week 1-4
- Scaling and Verifying Concurrent Users: Test and verify the system's ability to handle multiple users.
- Rate Limiting: Implement rate limiting to manage system load.
- Multi-Language Support - Batching: Develop support for multiple languages and optimize through batching.
- Immersive Voice Mode: Create an immersive voice mode for various applications.
- Fine-Tuning Models: Continuously fine-tune the models based on performance data.
- Automated Red Teaming: Simulate attacks to identify and fix vulnerabilities.
- Community Work Plan: Engage with the community for feedback and support.
- Feature Requests and Pull Request Management: Manage feature requests and pull requests from the community.
- Fixed Schedule of Uptime and Test Plans: Establish a fixed schedule for uptime and test plans.
- 3rd Party Integration: Integrate with third-party services and platforms.
Month 3
Week 1-4
- Resource Maximization: Optimize resource usage for scalability.
- Performance Monitoring: Continuously monitor performance and make necessary adjustments.
- Beta User Release: Prepare for and execute the release to beta users.
- Weekly Progress Updates: Continue providing weekly updates on progress and cost metrics.
- Batch Optimization Framework: Develop a framework for batch optimization, focusing on lecture conversion and archival work.
- Dataset Creation - Opt-In: Create datasets through opt-in prompts in the app for selection.
- Mobile App - Setup for Voice Mode: Develop and set up a mobile app for voice mode.
Todo MLOps
Observe speed of inference
Build online measurable document
Make app - production grade
Stres test - provide fast failure and feeback
More than 15 secs / Fail fast for unpaid / unlogged users
Build ddos / ip- tracking for load testing
Netflix / perplexity style building of feature and release
Build demo examples / jupyter notebook
Api key - bearer key management
User management with fastapi and react/ material ui
Federated learning -
Caching for tts
Langfuse/posthog
Add analytics for all services