Manas Peshwe
Download
Recommend
Messages
Share
M

Manas Peshwe

About

Detail

New York, United States

Timeline

work
Job
school
Education

Résumé

Jobs verified_user 0% verified
  • Dgraph Labs
    Software Engineer
    Dgraph Labs
    Sep 2022 - Jul 2024 (1 year 11 months)
    • Developed and deployed domain-specific RAG chatbot using generative AI and embedding models, improving customer support outcomes by 35% through comprehensive evaluation strategies and testing frameworks. • Built end-to-end RAG workflows with OCR, PII masking, vector indexing, and LangChain orchestration, improving response times across production AWS Bedrock environments using real-time data processing. • Designed and implemented scalable APIs with containerization using Docker to streamline data pipelines and integrate with LLM workflows, ensuring robust performance in enterprise-grade cloud infrastructure systems. • Implemented comprehensive AI model validation frameworks with automated testing pipelines, achieving 50% reduction in
  • Tata Consultancy Services
    System Engineer
    Tata Consultancy Services
    Jul 2021 - Aug 2022 (1 year 2 months)
    • Developed and deployed predictive models and ML algorithms on AWS (Lambda, SageMaker, DynamoDB), delivering scalable cloud infrastructure for enterprise financial analytics with data visualization dashboards. • Built end-to-end MLOps CI/CD pipelines using Python with Kubernetes orchestration, cutting model training and deploy ment times by 30% across 15+ production workflows. • Automated document processing with AWS Textract/Comprehend and integrated APIs with Apache Kafka for seamless real-time data flow, reducing manual workload by 80% and accelerating compliance reporting in enterprise systems. • Designed and implemented secure data pipelines with encryption and access controls using graph databases, ensuring com pliance with finan
Education verified_user 0% verified
  • University at Buffalo
    Master of Science
    University at Buffalo
    Aug 2024 - Current (1 year 4 months)
Projects (professional or personal) verified_user 0% verified
  • M
    Multi-Model Training System with GRPO & Multi-View Fine-Tuning
    • Built distributed training system combining 1.5B student, 7B reward model, and 14B meta-teacher using PyTorch and SGLang, deployed on dual A100 GPUs with containerization. • Implemented Group Relative Policy Optimization (GRPO) with adaptive feedback loops and rigorous evaluation method ologies, achieving 20% relative improvement on GSM8K compared to baseline LLM fine-tuning using generative AI techniques. • Engineered 4-level adaptive scaffolding (Heavy, Medium, Light, Minimal), dynamically adjusting based on training progress and reducing failures by 50% through real-time data processing.
  • C
    Consensus-Based Key-Value Store with Fault Tolerance
    • Implemented Raft consensus algorithm in Go with automated leader election, log replication, and recovery mechanisms, ensuring state machine consistency across 5-node distributed system for high-availability applications using cloud infrastructure. • Built high-performance key-value store achieving 12,000+ ops/sec throughput with 50ms failover during simulated crashes using heartbeat and quorum commit logic in production-grade environment with Docker deployment. • Designed intelligent client request routing with leader redirection and queueing mechanisms, ensuring seamless operation during leader transitions and reducing client timeout errors by 80% through data pipelines optimization. • Developed comprehensive testing framework simula