TAKESHI YAMAMOTO

Available to hire

Senior AI Engineer with 11 years of experience building production backend systems and AI products.

About

Senior AI Engineer with 11 years of experience building production backend systems and AI products. Focused on LLM workflows, structured outputs, RAG/model integration, evaluation, and reliability under ambiguous requirements, with strong debugging across application logic, data, model behavior, and user-facing workflows.

Experienced delivering both commercial and self-hosted model deployments, improving latency and cost via routing, caching, batching, and inference optimization. Known for failure-resistant API design, traceability, and test/evaluation loops that prevent regressions before release.

  • Joined August 2026

Languages

  • JavaneseFluent
  • EnglishAdvanced
  • EstonianBeginner

Industry experience

  • Software & Internet
  • Computers & Electronics
  • Professional Services

More experts in Tartu, Estonia

Projects · 1

• Built an LLM documentation workflow that transformed caregiver notes and voice transcripts into structured care records, summaries, and follow-up tasks using Python, FastAPI, and schema-constrained generation.
• Implemented Pydantic and JSON Schema validation with automatic repair and validation-driven human-review routing for incomplete, contradictory, or malformed model outputs.
• Developed RAG over care-plan context plus an evaluation suite covering unsupported claims, missing fields, invalid dates, and contradictory outputs; instrumented retrieval, inference, validation, and latency with OpenTelemetry.
• Added provider-neutral support for commercial and open-source models with Redis-backed context caching, allowing model changes without coupling the application workflow to one inference provider.

AI Clinical Documentation Assistant
  • AI Engineer
  • DevOps Developer
  • Full Stack Developer

Work experience· 6 roles

Senior AI Engineer · BPW ConsultingMay 2025 – Jul 2026 · 1 yr 3 mos

Turned ambiguous product concepts into production LLM features using Python and FastAPI: defined API boundaries, structured outputs, validation rules, and fallback paths, then rolled out via user-facing workflows. Built evaluation loops using curated examples, schema checks, semantic quality scoring, and regression cases to catch prompt/model/parsing regressions before release. Debugged AI feature failures spanning malformed inputs, retrieval context, model responses, backend state, and UI-visible behavior; added traceable intermediate outputs and failure classifications to speed diagnosis. Improved a latency-sensitive LLM workflow by introducing response caching, concurrent retrieval, timeout budgets, and model routing, reducing median test latency by ~35% while preserving evaluation quality. Used Claude Code, Cursor, and GitHub Copilot daily to implement, generate tests, refactor, and debug, verifying changes against application behavior and production constraints.

AI Platform Engineer · Breakwater TechnologyNov 2023 – Apr 2025 · 1 yr 6 mos

Integrated commercial and open-source LLMs behind provider-neutral Python adapters while keeping application contracts stable to enable model changes based on quality, latency, cost, and operational needs. Implemented self-hosted inference with vLLM and containerized model serving, configuring batching, memory limits, health checks, and request queues. Designed extraction workflows that converted unstructured text into typed JSON using schema-constrained generation, deterministic validation, repair attempts, and human review routing. Added bounded retries, circuit breaking, timeouts, and fallback models around external inference dependencies so partial provider failures degraded individual requests rather than blocking entire workflows. Built Pytest-based integration suites covering invalid payloads, unavailable models, malformed structured responses, and database failures. Profiled token usage and queue delay to identify cost/latency bottlenecks and adjusted model selection and batchi

Machine Learning Engineer · Sensor FieldsMay 2018 – Oct 2023 · 5 yrs 6 mos

Built Python inference services separating feature preparation, model execution, and API delivery so model versions could change without coupling application releases to training workflows. Developed asynchronous data pipelines with PostgreSQL and message queues, validating schemas and isolating malformed records before downstream prediction/automation. Containerized inference workloads with Docker and Kubernetes, defining resource limits, readiness checks, and safer rolling deployments. Reduced median benchmark latency by ~30% via profiling serialization, database access, and inference execution, then adding batching and Redis-backed caching. Created regression tests around feature transformations, model-version transitions, invalid input, and dependency failures. Instrumented services with OpenTelemetry metrics and traces for request latency, error classes, queue depth, and resource saturation to improve production diagnostics. Reviewed designs, guided engineers on service decomposit

Backend Engineer · Rail Baltic EstoniaJul 2016 – Apr 2018 · 1 yr 10 mos
Backend Engineer · Rail BalticJul 2016 – Apr 2018 · 1 yr 10 mos
Software Engineer · AdcashJun 2015 – Jun 2016 · 1 yr 1 mo

Education

Bachelor of Science in Computer Science · University of Tartu2010 – 2015
Bachelor of Science in Computer Science · University of Tartu2010 – 2015
Bachelor of Science in Computer Science · University of Tartu2010 – 2015

Hire TAKESHI YAMAMOTO today

Post a job and invite TAKESHI, or contact them directly. No hidden commission.

Post a job