Aayush Yadav
Software engineer — Tokyo, Japan
I build retrieval and LLM systems that reach production — for public-sector and construction clients in Japan, in Japanese and English.
- Based
- Tokyo, Japan
- Focus
- RAG · LLM · full-stack
- Japanese
- JLPT N2
- Available
- December 2026
About
AI-focused software engineer in Tokyo. Two years building RAG and LLM systems for Japanese public-sector and construction clients.
- RAG and LLM systems in production, not demos (done)
- Retrieval pipelines — chunking, hybrid search, reranking (done)
- Document extraction from construction and tender PDFs (done)
- Geospatial data applications and map visualisation (done)
- Mostly Python and TypeScript (done)
- Requirements definition through design, build and test (done)
- Release and maintenance — including whatever breaks after launch (done)
- Full-stack when the project needs it (done)
- Working directly with client stakeholders (done)
- Agile teams of four to six (done)
- English — native (done)
- Hindi — native (done)
- Japanese — professional working proficiency (JLPT N2, Dec 2025) (done)
- Spec discussions run in both English and Japanese (done)
- Multicultural teams, on-site with Japanese clients (done)
- Permanent in-house engineering role in Tokyo (done)
- Full-stack, applied AI, or cloud (done)
- Available from December 2026 (done)
- Currently in Japan on a valid work visa (done)
Experience
Five roles in about three years, most recent first — from a remote internship to on-site delivery for Japanese public-sector clients.
- 001
Pacific Consultants — Software Engineer
JUN 2025 – PRESENT[IN PROGRESS]- DX projects for construction and public infrastructure (done)
- Built the team's first production RAG assistant (done)
- Requirements definition through to maintenance (done)
- Client meetings conducted in Japanese (done)
- 002
Suntec — Frontend Engineer
FEB – MAY 2025[DONE]- Fleet and vehicle data platform, on-site Nagoya (done)
- Driver and operator modules, auth flows, dashboards (done)
- Real-time vehicle views in Apache ECharts (done)
- Video playback and file transfer over AWS Amplify and S3 (done)
- 003
Japanese language training → relocation to Japan
JAN 2024 – JAN 2025[DONE]- Run in parallel with the internship, remote from India (done)
- Relocated to Japan on a work visa (done)
- JLPT N2 (Dec 2025) (done)
- 004
Human Resocia — Software Engineer Intern
JAN – JUN 2024[DONE]- Xreport invoice and reporting package, feature extension (done)
- Frontend ~60%: plan management, dashboards, auth (done)
- Backend ~40%: Excel and PDF report generation (done)
- React, Django, Node.js, MySQL (done)
- 005
Hevonic.AI — Software Developer Intern
JUL 2023 – JAN 2024[DONE]- AI interview preparation platform, used at NYU (done)
- Built the cover letter and resume generator end to end (done)
- Caching, SEO and image work to cut page loads (done)
- Next.js, Redis, MongoDB, AWS (done)
Skills
What I have shipped with, grouped by what it is for. No self-rated percentages — the projects are the evidence.
Frontend
What runs in the browser
TypeScript · JavaScript · React · Next.js · Vue.js · Vite · Tailwind CSS · MapLibre · Mapbox · Apache ECharts
API & Services
Between the browser and the model
Python · Java · C++ · FastAPI · Node.js · Express.js · Django · Flask · Spring Boot · REST APIs · Async and multithreaded processing
AI & Retrieval
Where the accuracy work actually is
Retrieval-Augmented Generation · Large language models · LangChain · OpenAI API · Prompt engineering · Function calling and tool use · Embeddings · Chunking and reranking · Semantic and hybrid vector search · pgvector · Index tuning · Retrieval evaluation · Retrieval pipeline optimisation · OCR pipelines · Self-hosted and local models
Data
What the answers are grounded in
SQL · PostgreSQL · MySQL · MongoDB · Redis · PostGIS · QGIS · pandas · NumPy · PyMuPDF
Cloud & Delivery
How it reaches production and stays there
AWS · Microsoft Azure · Docker · GitHub Actions · CI/CD · DigitalOcean · Vercel · Cloudflare · Bash
- AWS Certified Cloud Practitioner (2023)
- JLPT N2 (2025)
- Ethnus MERN Full Stack (2023)
- hours / year
- 1,400hours / yearof manual document work removed by retrieval tuning alone.
- months
- 2monthsfrom sandbox to daily production use, over ~400,000 documents.
- hour, was days
- <1hour, was daysto prepare a deck, once the generator was in.
- latency, perceived
- 40%latency, perceivedcut by token-level streaming in the chat assistant.
Projects
Six builds, led by the ones with numbers attached. Most are client work delivered on-site in Tokyo.
Production RAG assistant
Pacific Consultants 株式会社
LangChain and LLMs over roughly 400,000 construction and tender documents, covering search, summarisation and Q&A. Out of the sandbox and into daily use in about two months; retrieval tuning alone took roughly 1,400 hours of manual document work off the year. Python and Express behind it, React and TypeScript in front, Docker out to DigitalOcean with Cloudflare ahead of it.
“Most of the accuracy work turned out to be retrieval, not prompting — chunking strategy, hybrid search over pgvector, and a reranking pass, measured against an eval set built from real user questions.”
- Python
- Express
- React
- TypeScript
- Docker
- DigitalOcean
- Cloudflare
- 002
AI presentation generator
MLIT / 国総研
Azure Web Apps · Functions · Application Gateway · VNet
- 003
MapAI
Otaru Canal · Kashiwa City
MapLibre · PostGIS · QGIS · Azure · Neon
- 004
Tender PDF extraction pipeline
PyMuPDF · Multithreaded · CSV / Excel
- 005
ManiKani
Spaced-repetition learning SaaS
Next.js · FastAPI · PostgreSQL · pgvector · Redis · Stripe
- 006
AI chat assistant
Streaming · Prompt templates · Rate limiting
Contact
Open to permanent in-house engineering roles in Tokyo from December 2026.
- GitHub
- github.com/AayushDev-02
- Location
- Tokyo, Japan
- Available
- From December 2026
- Documents
Resume and 職務経歴書 available on request, along with a 履歴書 if your process needs one.