Conversational AI Platform - Austria
Principal AI Engineer, Mar 2023 – Present
Project description
- Led the architecture and deployment of an enterprise-grade conversational platform using OpenAI's GPT-4 and LangChain. Enabled real-time customer interaction at scale for over 50 enterprise clients.
Responsibilities
- Architected a modular, multi-tenant Conversational AI Platform using LangChain and OpenAI's GPT-4, enabling dozens of enterprise customers to deploy custom chat experiences with minimal engineering effort.
- Designed a dynamic agent routing framework where conversations are steered based on user intents, tool availability, and organizational context.
- Integrated vector databases (Pinecone) for domain-specific document retrieval, enhancing chatbot grounding and reducing hallucinations.
- Built internal developer tools to streamline prompt iteration, chain design, and debugging, allowing rapid prototyping of new LLM-powered workflows.
- Conducted internal training sessions and peer reviews for junior engineers; co-authored internal documentation for best practices in LLM safety, token efficiency, and cost optimization.
- Developed observability stack with Prometheus, Grafana, and custom OpenAI token usage analytics; reduced monthly LLM usage costs by 28%.
Technologies
- Python, OpenAI GPT-4, LangChain, Pinecone, FastAPI, Prometheus, Grafana, GitHub Actions, Docker, Kubernetes, AWS (S3, Lambda, EC2).
LLM-powered Support Bot - Switzerland
Senior AI Engineer, Aug 2020 – Dec 2022
Project description
- Developed a support chatbot using GPT-3.5 to handle over 80% of tier-1 queries in a SaaS product used by 1M+ users.
Responsibilities
- Built and productionized a customer support chatbot powered by GPT-3.5 that resolved 80%+ of user queries without human intervention.
- Fine-tuned the model using over 50K historical chat logs, combined with retrieval-augmented prompts to improve contextual relevance.
- Designed robust fallback and recovery mechanisms, including rephrasing suggestions, clarifying questions, and escalation triggers.
- Integrated the bot with Zendesk, enabling real-time ticket creation, prioritization, and tagging based on detected sentiment and urgency.
- Created a feedback loop that logs low-confidence responses and feeds them back into weekly training pipelines for continual improvement.
- Led a cross-functional team in conducting A/B experiments to validate hypothesis-driven prompt variants, improving user satisfaction (CSAT) by 30% and reducing first response time by 60%
Technologies
- Python, GPT-3.5, Hugging Face Transformers, LangChain, RAG, Redis, FastAPI, Azure OpenAI, Zendesk APIs, MLflow, Kubernetes.
E-commerce (Recommendation) - Singapore
AI Developer, Mar 2018 – Jul 2020
Project description
- Built ML-powered recommendation engines and personalized search systems for a large online retail platform..
Responsibilities
- Developed scalable personalized recommendation engines using collaborative filtering, hybrid models, and deep learning architectures.
- Implemented real-time inference APIs for delivering context-aware product suggestions based on browsing patterns and purchase history.
- Built feature engineering pipelines to encode product taxonomy, user embeddings, and event stream data using Spark and Airflow.
- Worked with the marketing team to launch targeted campaigns using model insights, improving click-through rate (CTR) by 35% and increasing average order value (AOV) by 22%.
- Delivered model evaluation dashboards with metrics like precision@k, recall, and diversity scores; automated nightly retraining jobs based on fresh behavioral data.
- Collaborated with data scientists and product managers to define new experiments and optimize end-to-end user journeys.
Technologies
- Python, Scikit-learn, XGBoost, TensorFlow, Spark, Airflow, PostgreSQL, Redis, FastAPI, Docker, AWS (S3, SageMaker)
Smart Document Processing - Singapore
AI Developer, Jan 2016 – Feb 2018
Project description
- Created intelligent recommendation engines for a recruiting startup, matching candidates with jobs based on skill semantics.
Responsibilities
- Created a robust document understanding system capable of parsing scanned invoices, tax forms, and contracts using deep learning and OCR.
- Engineered an ensemble of CNN-CRNN networks and OpenCV pre-processing modules for layout analysis, text extraction, and noise reduction.
- Trained BERT-based models for document classification and field tagging, replacing manual labor in data entry with over 94% accuracy.
- Designed a distributed processing architecture using Docker, AWS S3/EC2, and Celery for large-scale batch extraction of financial documents.
- Delivered a document analytics dashboard for finance teams to track processing status, exception handling, and SLA adherence.
- Reduced manual verification time by over 70%, freeing up headcount for higher-value analysis work.
Technologies
- Python, OpenCV, Tesseract OCR, CNN, CRNN, BERT, TF-IDF, XGBoost, Flask, Celery, AWS EC2, S3, Docker
