Machine Learning Engineer — GenAI · Production AI

I build AI systems where accuracy, latency, and cost have to be right at the same time. RAG, agents, and LLM apps that survive contact with real users.

Bengaluru, India · b.lourdhuraju1234@gmail.com · LinkedIn · Kaggle

Machine Learning Engineer · Sujanix Pvt. Ltd. · Bengaluru — Jan 2026 – present

Production OCR at scale for government utility automation. Architected Transformer-based pipelines on SVRT (a Swin-V2-based regression transformer); improved exact-match accuracy by 82.74% to 97.04% via a custom FocalCTCLoss addressing class imbalance. Deployed via INT8 ONNX on AWS Lambda + FastAPI — 30% inference-cost reduction at 99.9% uptime.

Autonomous agentic AI on LangGraph with a three-tier memory architecture (Episodic · Semantic · Procedural) for long-horizon reasoning. CAT/IRT psychometrics with Fisher Information maximization reduced assessment time 70% (30 min → 9 min). Privacy-first local deployment; XTTSv2 zero-shot voice cloning.

LangGraph · CAT/IRT · Fisher Information · XTTSv2 · 3-tier memory

Production financial document intelligence for regulated customers. Local LLMs via Ollama with ChromaDB for full data sovereignty. JWT-based multi-tenant session isolation. Multi-modal extraction from PDFs and bank statements via local VLMs.

Ollama · ChromaDB · VLM · JWT multi-tenant

97% exact-match accuracy on numeric meter datasets via an SVTR backbone, custom Feature Rearrangement Modules, and FocalCTCLoss. ONNX INT8 quantization for edge deployment. Custom Semantic Guidance Modules for robustness in motion-blur and low-contrast environments.

SVTR · FocalCTCLoss · ONNX INT8 · FRM

Sujanix Pvt. Ltd. — ML Engineer · Jan 2026 – present Production OCR pipelines, scalable inference, end-to-end model reliability.

SpaceDrift — Founder · Aug 2024 – Dec 2025 MSME-registered sole proprietorship. Delivered engineering builds, data annotation services, and research support to PhD scholars (problem framing, dataset curation, experimental pipelines). Engaged paid contractors on a per-project basis when workload exceeded solo capacity.

BrainOvision Solutions — Data Science Intern · Feb 2024 – Apr 2024 Improved forecasting accuracy by 15% through ensemble gradient boosting and structured feature engineering.

Core — Python · SQL · C++ ML/DL — PyTorch · TensorFlow · scikit-learn · Hugging Face · OpenCV · XGBoost GenAI/LLM — LangChain · LangGraph · Ollama · vLLM · Groq · OpenAI Retrieval — ChromaDB · Qdrant · Pinecone · sentence-transformers Serving & Infra — FastAPI · Docker · Kubernetes · AWS (EC2 · Lambda · S3 · SageMaker) · ONNX · TensorRT Data — PostgreSQL · MongoDB · Redis MLOps — MLflow · Weights & Biases · DVC · Git · Linux

• Kaggle Expert (Notebooks tier) — deep learning competitions
• Machine Learning Specialization — DeepLearning.AI / Stanford
• Data Science using Python — NPTEL / IIT Madras

• Production RAG evaluation framework — public benchmark + leaderboard
• Agent benchmark for Indian-language tasks
• Open-source contributions to vLLM, TRL, LlamaIndex (2026 H2)

Open to roles and collaborations in applied ML, agentic AI, RAG systems, LLM inference, and production GenAI.

b.lourdhuraju1234@gmail.com · LinkedIn