Pratibha Kambi

ML/AI Engineer

ML/AI Engineer with 4.5 years of production software engineering experience at Bosch and hands-on GenAI project delivery through Omdena, currently completing an M.S. in Applied Artificial Intelligence. Built an agentic, multilingual RAG assistant (LangGraph orchestration, hybrid retrieval, rigorous IR/LLM-judge evaluation) and a production-style XGBoost forecasting pipeline deployed and monitored on AWS SageMaker. Combines applied ML/GenAI system design with the production- engineering discipline (CI/CD, root-cause analysis, testing rigor) of 4+ years shipping automotive software at enterprise scale.

pratibhakambi@gmail.com+91-8296910231linkedin.com/in/pratibha-kambi-8492022a6github.com/pratibhabd

Skills

RAG pipelinesAgentic AI (LangGraph)Prompt EngineeringLLM-as-Judge evaluationMulti-agent orchestrationRegressionClassificationTime-Series ForecastingXGBoostFeature EngineeringDeep & Shallow Neural NetworksNLPXLM-RoBERTaWhisper (ASR)AWS SageMakerS3AthenaModel Monitoring (drift, PSI, RMSE)Amazon BedrockPythonC++StreamlitFastAPIGit/GerritPyCharmTableauVector Databases (ChromaDB, Supabase/pgvector)

Experience

AI/ML Engineer (Project-Based)

Omdena

2026 – Present

  • Developed a multilingual voice-first AI assistant for Bhutan public services, enabling citizens to access permit and business information through natural language voice interactions.
  • Designed a layered agentic workflow in LangGraph orchestrating Speech-to-Text, guardrails, intent classification, a ReAct orchestrator, and Text-to-Speech nodes for end-to-end request handling.
  • Built a hybrid intent classification and semantic routing pipeline (XLM-RoBERTa + Gemini LLM + rule-based guardrails), routing citizen queries across permit, business, general, and safety categories with [ADD METRIC: e.g., accuracy/F1 score].
  • Built a RAG pipeline using hierarchical document chunking, sentence-transformer embeddings, and Supabase (pgvector) hybrid retrieval (semantic + full-text) to ground responses in Bhutan public-service knowledge bases, achieving [ADD METRIC: e.g., Hit@K / Precision@K score].
  • Designed Supabase-backed conversation memory (user profile, chat history, decision logs) for persistent context and full workflow traceability.
  • Implemented secure PII handling, isolating sensitive citizen data (e.g., CID) from LLM prompts and conversation history via backend API calls made only when required.
  • Integrated Whisper-based ASR and multilingual translation to support Dzongkha–English voice interactions.
  • Evaluated the system end-to-end using Accuracy, Precision, Recall, Macro F1, Precision@K, Hit@K, and LLM-as-a-Judge relevance scoring for both intent routing and retrieval quality.
  • Built simulated government service integrations (FastAPI) for permit/registration workflows with structured responses and decision-trace logging.

Software Engineer

Bosch Global Software Technologies (BGSW)

2019 – 2024

  • Designed an ML-enhanced Automotive Guidance Proposal (AGP): built destination-prediction logic using historical trip patterns to personalize route guidance.
  • Built an XGBoost-based destination prediction pipeline using route frequency, origin, time-of-day, and day-of-week features to rank the most probable destination among habitual routes, reaching [ADD METRIC: e.g., top-1/top-3 prediction accuracy].
  • Proposed a hybrid architecture combining rule-based AGP eligibility logic with ML-based destination ranking, improving route proposal relevance while preserving explainability of navigation decisions.
  • Participated in BCAI (Bosch Center for AI) learning initiatives covering the end-to-end ML lifecycle — data preparation, model training, evaluation, and experimentation.
  • Analyzed large-scale production issues across middleware and core navigation systems (Python, C++), performing root-cause analysis on route guidance workflows and reducing time-to-diagnosis via custom log-tracing plugins.
  • Built and maintained automated regression and unit-testing pipelines, improving defect detection and reliability of navigation features.
  • Monitored CI/CD pipelines in Jenkins, investigating build failures and test regressions to protect code coverage and build stability.
  • Performed telemetry and system-log analysis to monitor runtime behavior and improve observability across distributed navigation components.
  • Collaborated cross-functionally to integrate and validate middleware components in large-scale automotive systems, with hands-on on-target testing and embedded debugging.
  • Used Git/Gerrit and GitHub for version control and code review within a large enterprise codebase.

Projects

◎

Retail Demand Forecasting for Inventory Optimization

Built an end-to-end time-series forecasting system predicting daily retail demand per store–product combination, with lag features, rolling statistics, and seasonality features.

  • Trained and evaluated XGBoost models, improving forecast accuracy over baseline by [ADD METRIC: e.g., X% lower RMSE/MAPE].
  • Deployed the model on AWS SageMaker for scalable batch inference, with a feature pipeline built on S3 and Athena.
  • Implemented SageMaker Model Monitor to track data drift, prediction drift, and performance degradation using RMSE and Population Stability Index (PSI).
◎

Explain My Prescription

Built a privacy-preserving, offline assistant using MedGemma (HAI-DEF healthcare LLM) to generate safe, non-diagnostic, patient-friendly medication explanations.

  • Implemented prompt-level guardrails restricting outputs to educational content, avoiding diagnosis or treatment recommendations.
  • Evaluated model outputs with a rule-based pipeline assessing safety, instruction-following, and readability.
  • Tech stack: LLM, Prompt Engineering, Hugging Face, Responsible AI techniques.

Education

University of San Diego

KLE Technological University, Hubballi

Certifications

  • The Complete Python Developer — Udemy
  • Natural Language Processing with Classification and Vector Spaces — Coursera
  • Neural Networks and Deep Learning — Coursera

Career highlights

January 2026

  • Started: AI/ML Engineer (Project-Based) at Omdena

    Omdena

January 2019

  • Started: Software Engineer at Bosch Global Software Technologies (BGSW)

    Bosch Global Software Technologies (BGSW)

Built with Round Zero