ML/AI Solution Architect at Dave AI (2025-05 – 2026-02)
Architected and built end-to-end conversational AI systems with production reliability patterns and multi-language support.
- Built an end-to-end conversational AI system connecting S2S models with telephony platforms (Twilio and Tata Tele) for real-time voice communication, maintaining 95% uptime
- Created scalable backend with 10+ REST API endpoints for ML model serving, implementing validation, error handling, and structured logging
- Cut TTS and ASR processing time from 3+ seconds to 1-2 seconds by optimizing model pipelines and backend workflows
- Implemented production reliability patterns (connection pooling, circuit breakers, retry logic) ensuring system stability during failures
- Developed Gryd-based caching layer reducing query response times from seconds to milliseconds and backend load by 40%
- Split "Synthesis Main" into GPU and non-GPU microservices, enabling independent scaling and reducing compute costs by 25%
- Set up audit logging and cost tracking to monitor model performance, API errors, and spending in real-time
- Shipped ASR, S2S, Translation, and TTS models supporting 10+ languages including multiple Indian accents
- Gave 3+ technical demos to clients and wrote documentation (API specs, deployment guides) for team onboarding
ML Intern at Dave AI (2024-08 – 2025-05)
Contributed to speech AI systems development with focus on framework migration and language model expansion.
- Led migration from Celery to in-house Gryd framework, enabling consistent task execution across local, UAT, and production environments
- Improved task management and API efficiency, boosting system reliability by 30%
- Added emotion support to TTS models and expanded language support to 10+ Indian languages and accents
- Built Gryd tasks automating natural language to SQL query conversion for Data Insights project
- Created benchmarking framework for comparing and selecting data insight models
- Set up Docker environments and database-driven config management for dynamic model fetching
- Worked with senior engineers on CI/CD pipelines, model optimization, and technical documentation
