Skip to content

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