3+ Years
12+ Projects
8 Languages

[ SYSTEM ONLINE ]

SAHIL CHAVAN

AI Research Engineer

Building Intelligent Systems at the Intersection of Speech AI & Gaming

START MISSION
ACHIEVEMENTS 1/7

[ MISSION BRIEFING ]

About The Operative

AI Research Engineer with 3+ years of experience specializing in Speech AI, Language Identification, and Gaming AI.

Expert in fine-tuning large-scale speech models across 8 Indic languages, building production-ready wake-word systems, and developing AI agents for FPS gaming.

Passionate about creating intelligent systems that bridge the gap between cutting-edge AI research and real-world applications.

3+ Years Experience
12+ Projects Delivered
8 Languages Supported

[ TECH ARSENAL ]

AI & Machine Learning

Speech & Audio

Frameworks & Tools

Cloud & DevOps

[ MISSION HISTORY ]

Combat Experience

  • Fine-tuned LID models (Facebook MMS, ECAPA-TDNN) for 8 Indic languages; improved robustness under noisy, real-world audio.
  • Designed and deployed wake-word detection (wav2vec 2.0) for “Hello Jio” — low-latency, always-on inference.
  • Implemented Silero VAD with ONNX runtime and open-source pip; benchmarked accuracy and inference.
  • Built Data Flywheel MLOps pipeline: ingest, store, version language-wise speech data; classify noisy/silent/clean; continuous retraining and evaluation.
  • Built Data Flywheel pipeline (Grafana Loki) for TTS, STT, Intelligence/LLM logs; generated tabular datasets (AI I/O, token usage, latency) for LID, Wake-Word, Whisper, Parakeet.
  • LLM evaluation and benchmarking (Perplexity, GPT-4, EXA AI, LinkUp AI) — latency, caching, query-response consistency.
  • Improved AI agent performance by 30% through dataset optimization and algorithm tuning.
  • Reduced AI training time from 8 hours to 4.5 hours (43%) via optimized metadata and preprocessing pipelines.
  • Developed AI-driven FPS gaming agents using deep learning and behavioral cloning.
  • Applied Computer Vision (OpenCV) for real-time perception and decision-making in FPS bots.
  • Collaborated in a 4-member AI team; reduced gameplay bugs by 20% via adaptive decision logic.
  • Built metadata collection and labeling systems to accelerate experimentation cycles.
  • Supported continuous improvements in AI agent stability and realism.

[ ARSENAL ]

Project Showcase

Speech AI

Language Identification System

Fine-tuned MMS and ECAPA-TDNN models for 8 Indic languages, addressing accent variation, background noise, and code-mixed speech.

Speech AI

Wake-Word Detection

Built a wav2vec-based wake-word system optimized for low false positives and low latency for the Hello Jio phrase.

AI Evaluation

LLM Evaluation Framework

Designed evaluation workflows for perplexity, response relevance, and caching efficiency across multiple LLM providers.

Gaming AI

FPS Gaming AI Agents

Developed AI-driven FPS gaming agents using deep learning and behavioral cloning for realistic gameplay.

MLOps

Data Flywheel Pipeline

Built MLOps pipeline for automatic data ingestion, versioning, and continuous model retraining using Grafana Loki.

Speech AI

Voice Activity Detection

Implemented Silero VAD using ONNX runtime, benchmarking accuracy and inference performance.

[ ESTABLISH CONNECTION ]

Initialize Contact

[ COMMUNICATION CHANNELS ]

sahildchavan04@gmail.com +91 9834020445

Pune, India

Land on the grid to unlock your first achievement.

ACHIEVEMENTS 1/7