You're seeing this page as if you were . The main menu is still yours, though. Exit from immersion
Malt Strategy
New report
Malt Tech Trends 2026
Sandeep NayakSN

Sandeep Nayak

AI Engineer

€175/day
Tokyo, JP
3-7 years
Python
AI Agent
MLOps
RAG
artificial intelligence

Average response time: 1 hour

About Sandeep

I build production-grade AI systems that solve real business problems. With 5+ years of experience across startups and enterprise in Japan, I specialize in LLM-powered agents, RAG pipelines, and MLOps — delivering measurable outcomes like 40% efficiency gains and ¥15M/year in cost savings.

Currently based in Tokyo, I've shipped AI platforms at Gaudiy (Web3/AI startup) and Rakuten Mobile. Whether you need a conversational analytics tool, a multi-agent workflow, a GPU inference platform, or an end-to-end LLM evaluation framework — I take ideas from prototype to production.

I work fully remote and communicate fluently in English and conversational Japanese.
Digital and IT Services
Telecommunications

  • English

    Native or bilingual

  • Japanese

    Conversational

Remote only
Primarily works remotely

Experience

  • Gaudiy Inc
    AI Engineer
    DIGITAL AND IT
    April 2024 - Today (2 years and 3 months)
    Japan
    1. BI Agent Platform
    • • Supported a 10B yen investment story by delivering a production BI conversational tool with Hybrid RAG over millions of MAL user records, integrating BigQuery, Elasticsearch, and a vector database. (BigQuery, Elasticsearch, Hybrid RAG, GraphRAG)
    • • Reduced topic fragmentation and improved categorical consistency by building a hybrid vector+fuzzy tag consolidation agent that deduplicates LLM-extracted topics into canonical themes using similarity thresholds and LLM arbitration. (pgvector, NLP, LangChain)
    • • Built a custom LLM-driven iterative prompt optimization algorithm for a 5-class relevance annotation problem —benchmarked against DSPy (BootstrapFewShot, MIPRO) with Macro-F1 as optimization target and per-class degradation guardrails. (Python, DSPy, LLM Evaluation, LangChain)
    • • Reduced embedding retrieval cost by researching 13 models on a custom Japanese conversation dataset; identified local multilingual models as cost-effective alternatives to cloud APIs and published findings as a bilingual research blog post. (Embeddings, Japanese NLP, Research, OSS)
    • • Strengthened hypothesis validation and downstream insight quality by building a real-time interview agent with voice and text interactions as a data-collection layer for the BI platform. (WebRTC, Multi-Agent, Voice AI)
    2. AI Platform & Observability Foundation
    • • Reduced debugging time approximately 30% by introducing end-to-end observability and traceability across multi-agent workflows. (LangSmith, Datadog)
    • • Enabled cost-efficient AI inference at scale by co-architecting a fully auto-scaling GPU inference platform on GKE with separate CPU/GPU node pools, CI/CD, Cloud Armor, and Terraform IaC. (GKE, Terraform, GitHub Actions, Kubernetes, Cloud Armor)
    • • Expanded external technical credibility and the broader fundraising narrative by open-sourcing langsmith-evaluation helper, a Python library for declarative LLM evaluation via YAML configs. (Python, LangSmith, OSS)
    Python artificial intelligence Langchain RAG MLOps
  • Rakuten Mobile
    Software Engineer (AI)
    TELECOMMUNICATIONS
    September 2022 - March 2024 (1 year and 6 months)
    Japan
    1. LLM Analytics & Knowledge Workflows
    • • Reduced query handling time 40% by developing a natural-query analytics agent for efficient analysis across millions of records. (LangChain, LLMs)
    • • Boosted text processing efficiency 80% by implementing RAG-based extraction and summarization workflows. (RAG, Hugging Face) 2. Customer & Network Optimization Models
    • • Improved retention and sales outcomes by leading a team of 3 to deliver customer propensity modeling for millions of users.
    (PySpark, Hadoop)
    • • Saved approximately 15M yen/year by deploying an end-to-end time-series DNN model for energy optimization.
    • • [Patent Applied] Saved 400 hrs/year/person by cutting ticket MTTR with 3 production AI systems for content-based similar ticket recommendation, smart ticket fill, and auto root-cause analysis. (PySpark, Kafka, Kubeflow)
    • • Improved production AI reliability by operationalizing model lifecycles across deployment, monitoring, and iteration for these systems. (Kubeflow, MLOps)
    Python MLOps Langchain RAG PySpark
  • Tata Motors
    Senior Manager (Planning)
    AUTOMOBILE
    August 2016 - September 2017 (1 year and 1 month)
    India
    1. Manufacturing Process Optimization
    • • Reduced process time 40% by using AI-driven value stream mapping to identify bottleneck parts in manufacturing workflows. (Python, Analytics)
    • • Enabled proactive maintenance by developing ML models for part-failure prediction during assembly processes. (Python, Scikit-Learn)
    • • Improved stakeholder decision-making by presenting insights that supported data-driven maintenance and process optimization. (Analytics, Stakeholder Communication)
    Python Scikit-learn Machine learning
Check out Sandeep's experience

Recommendations

Be the first to recommend Sandeep

Help this freelancer shine by sharing your experience working together.

These freelancer profiles also match your criteria

AgathaA

Agatha Frydrych

Backend Java Software Engineer

4.7

(3)

2

BaptisteB

Baptiste Duhen

Fullstack developer

4.6

(4)

5

AmedA

Amed Hamou

Senior Lead Developer

4

(2)

7

AudreyA

Audrey Champion

Web developer

4.3

(3)

4

Signup to reveal

Education

  • Applied ML Researcher
    Top National University
    2022
    Applied ML Researcher
  • Master's in Applied Information Science
    2021
    Master's in Applied Information Science
Check out Sandeep's education

Categories