AI Engineer focused on production GenAI systems, reliable LLM infrastructure, retrieval systems, and robustness evaluation under real-world constraints.
class SarthakChauhan:
role = [
"AI Engineer",
"ML Systems Builder",
"Research Engineer"
]
interests = [
"Reliable LLM Systems",
"Distributed Inference",
"Retrieval-Augmented Generation",
"Vision Robustness",
"Temporal Memory Systems",
"Evaluation Under Distribution Shift"
]
currently_building = [
"Production-scale GenAI infrastructure",
"Long-context retrieval and reranking systems",
"Low-latency async AI systems",
"Reliable LLM evaluation pipelines"
]
|
|
|
AI Safety RL Environment Finalist — Meta × Scaler PyTorch OpenEnv Hackathon (Top 2.6%)
Stack:
|
Memory System for Conversational AI
Focus Areas: Temporal reasoning • retrieval • reranking • memory systems Stack:
|
|
Real-Time Semantic Question Clustering
Infra: Redis workers • pgvector • WebSockets • circuit breakers Stack:
|
Production GenAI System @ Cograd
Infra: Async orchestration • Redis • Qdrant • Azure OpenAI Stack:
|
Evaluating 12 ImageNet-pretrained architectures across IN-Val, IN-V2, IN-R, IN-A, and IN-C using:
- ECE
- AURC
- selective prediction
- corruption robustness
- universal failure analysis
Benchmarked 10 dehazing architectures and identified a 15–20 dB PSNR gap between synthetic benchmarks and real dense-fog highway conditions.
Improved F1 from 0.784 → 0.866 on a 700K-post dataset using:
- XLM-R transfer learning
- BiGRU attention fusion
- multilingual representation learning
AICAPS 2026 — IEEE Kerala Section First Author
DICCT 2026 First Author
IC3SE 2025 — IEEE UP Section Second Author
- 🏆 Meta × Scaler PyTorch OpenEnv Hackathon — Finalist (Top 2.6%)
- 🏆 Amazon ML Challenge 2024 — Top 0.5%
- 🏆 IIT Bombay Convolve — Top 50 / 4189 teams
- 🎓 Dean’s List — Top 10%
- 📚 GPA: 9.42 / 10.0
Building reliable AI systems, retrieval infrastructure, and evaluation pipelines.
