Skip to content
SK

Saurabh Kumar

CTO & Co-Founder / Full-Stack AI Engineer

Tokyo, Japan

๐ŸŸขAping In

2x technical co-founder with 5+ years shipping end-to-end production AI systems. Built LLM pipelines with 75% cost reduction through intelligent caching, ML systems achieving 92%+ accuracy at scale, and platforms serving 200K+ users. Full-stack AI: Agentic Systems, Harness Engineering, fine-tuning, RAG, agents, computer vision, edge deployment. Led 10-person team from 0โ†’1. IEEE-published researcher.

Work Preferences

Salary

USD 150,000 โ€“ 200,000

Skills

Fine-tuning (LoRA/PEFT)9/10
HuggingFace Transformers9/10
Python9/10
LLM Pipelines / Prompt Engineering9/10
Machine Learning9/10
MLOps (DVC, W&B) (8/10)Computer Vision (8/10)Firebase / Firestore (8/10)GCP / Cloud Infrastructure (8/10)PyTorch (8/10)RAG / Vector Search (ChromaDB) (8/10)Flutter / Dart (8/10)TensorFlow / TFLite (8/10)TypeScript / Node.js (7/10)Edge Inference (OpenVINO / TFLite) (7/10)Go (7/10)gRPC / Protobuf (7/10)Kubernetes (7/10)Redis (7/10)Android (Kotlin) (7/10)

Work Experience

CTO & Co-Founder

Adora Inc.

Jul 2023 โ€” Present

Led technical development of Kodomamo, a child safety platform serving 200K+ users. Built unsafe website detection API with 4-layer LLM pipeline (P50: 5ms, 1000+ req/s), fine-tuned BERT/ShieldGemma for Japanese toxicity classification (92% accuracy), built AI code review agent analyzing 2300+ PRs, NSFW detection ML pipeline with 40% FP reduction, event-driven serverless GCP backend, LINE message monitoring system scaling to 1000+ concurrent sessions, and 10+ Flutter native plugins. Led 10-person engineering team.

Perception ML Engineer

Rapyuta Robotics

Jun 2021 โ€” Oct 2022

Built multi-model object detection system (RetinaNet, CenterNet, SSD+FPN) for autonomous warehouse robotics. Achieved 10X faster training via data pipeline optimization. Deployed edge inference using OpenVINO INT8 quantization on Intel NCS2 with ROS integration for real-time obstacle detection.

AI Consultant โ€“ Sports Analytics

MTX Group

Feb 2021 โ€” May 2021

Built Action Quality Assessment system using I3D for spatiotemporal feature extraction and LSTM with temporal attention for athletic performance scoring. Integrated pose estimation for biomechanical analysis with real-time coaching feedback pipeline.

Co-Founder & Tech Lead

Smartbooks Inc.

Jan 2021 โ€” Present

Built Japanese audiobook mobile platform. Developed internal audiobook production QA tool in Go with FFmpeg and GCP TTS, EPUB-to-audiobook ETL pipeline using parallel processing, and Flutter audiobook player with offline support, RevenueCat IAP, Algolia search, and Firebase integration.

Software Engineer Consultant โ€“ Network Analytics

Nokia

Dec 2019 โ€” Aug 2020

Built graph digitization system using LSTM-based OCR for network topology extraction from scanned engineering diagrams. Achieved 6X speedup via recursive image segmentation. Designed failure detection pipeline correlating topology with real-time network telemetry.

ML Intern โ€“ News/Ads Recommendation

Yahoo! JAPAN

May 2019 โ€” Jul 2019

Built demographic estimation model using DSSM + attention-BiLSTM on user browsing patterns, achieving 92% gender and 70% age prediction accuracy. Integrated model into news/ads personalization pipeline. Received Pre-Placement Offer.

Education

Indian Institute of Technology (IIT) Delhi

B.Tech ยท Electrical Engineering

2016 โ€“ 2020