AI/ML Engineer with three years of professional experience in deep learning, NLP, and generative AI. I build production ML systems end to end — recommendation engines, LLM and RAG pipelines, and computer vision applications — with hands-on MLOps across Docker, Kubernetes, and self-hosted GPU model serving.
Currently a Werkstudent AI/ML Engineer at LikeTik (Axinity GmbH), building production recommendation and content-analysis pipelines, while writing my Master's thesis at THWS on security failure propagation in multi-agent coding systems. Available for full-time roles.
Built an autonomous 5-node LangGraph pipeline that generates institutional-grade equity research reports, covering RAG over SEC 10-K/10-Q filings via Qdrant hybrid search, real-time news sentiment, and a self-correcting Critic Agent with confidence gating, reducing manual research from ~6 hours to ~60 seconds across 6 global exchanges.
Co-developed a reasoning-tuned Gemma-2B for the Google Tunix Hackathon (team of three), owning the inference pipeline. Fine-tuned with LoRA (rank 32, alpha 64) on ~570k samples from MetaMath, OpenThoughts, Medical-O1, Bespoke-Stratos, and GSM8K, with 4-bit NF4 quantization.
Fine-tuned ResNet18 for pneumonia detection from chest X-rays with Grad-CAM and LIME explainability. Compressed MobileNetV2 anomaly detector via structured pruning and INT8 quantization — model size reduced from 8.7 MB to 4.4 MB (−49%) with no accuracy loss.
Trained YOLOv8 detector on thermal UAV imagery for automated waterfowl detection. Achieved 86.44% mAP@0.5 and 93.21% Precision on 83 test images with 1,411 ground-truth annotations. Complete end-to-end CV pipeline from preprocessing to evaluation.
Trained and compared multiple RL agents (Q-Learning, Monte Carlo) converging towards the mathematically optimal Blackjack strategy. Results documented in a published technical report.
Built CNN classifier for real-time sign language gesture recognition across 24 gesture classes. Robust preprocessing pipeline with data augmentation for generalisation to unseen hand positions.
Real-time messaging app with bidirectional WebSocket communication, JWT authentication, and end-to-end encryption. Scalable room-based architecture deployed on self-managed Linux server with NGINX.
Optimal Strategy Learning in Blackjack using Deep Reinforcement Learning
Om Borda
2025
This paper explores the application of reinforcement learning algorithms to learn optimal Blackjack playing strategies. We implement and compare Q-learning and Monte Carlo methods, demonstrating convergence to near-optimal play after extensive training episodes.
Built a hybrid recommendation system matching ~1M supplier products to thousands of TikTok creators using multilingual embeddings, Qdrant vector search, metadata filtering, and an LLM re-ranking layer returning the top five product suggestions per creator
Built a video content-analysis pipeline using computer vision (Qwen-VL) for visual scene understanding and PyTesseract for on-frame text extraction, feeding signals back into the recommendation features
Handled MLOps end to end: Dockerised services on Hetzner with Kubernetes for scaling, FastAPI for the backend, self-hosted Qwen 7B served on Trooper.ai (A100 40GB), and PostgreSQL for metadata
Built a customer churn prediction pipeline (pandas, scikit-learn, XGBoost) with feature engineering on activity trends, session duration, failed-payment ratio, and support-ticket frequency, retrained weekly on fresh data
Developed a sales analytics and forecasting dashboard in Streamlit with KPI cards, regional heatmaps, regression-based monthly forecasts, and product-level breakdowns, consolidating data previously spread across Excel sheets and client databases
Full Stack Developer
Greendotslab Software Solutions
May 2022 – Jun 2023 · Gujarat, India
Built and deployed 10+ client web applications end-to-end using Node.js and React, with load balancing and modular architecture for production
