Kathmandu, Nepal • imbibek8366@gmail.com

I am a software-minded Machine Learning Engineer bridging the gap between mathematical theory and production-grade software systems. Currently working as an AI / ML Intern at FlyRank AI, engineering ML pipelines across 70M+ row data warehouses, building custom LLM components, and deploying asynchronous backends.

• Current Status: Coursework fully completed. Final BCA degree examinations begin on August 11, 2026 (lasting approx. 1–2 weeks). Available for immediate recruitment & full-time deployment starting mid-August 2026 upon exam completion (seeking Associate / Entry-Level Machine Learning Engineer, AI Engineer, or MLOps Engineer roles).
• Flexibility: My current internship is self-paced, and all coursework/training programs are complete, allowing 100% dedicated focus to full-time engineering commitments once exams conclude.
• Career Journey: Spent my university years balancing degree requirements alongside targeted contract software engineering roles, academic internships, and production ML pipelines.
• Tech Stack Focus: The tools listed below are strictly technologies I have actively shipped in production environments, deployed in microservices, or used to build open-source ML architecture. (Secondary exposure to languages like PHP, C#, Java, and C is intentionally omitted to highlight primary engineering depth).

• 🧠 LLMs & Core Mechanics: Tokenization engines from scratch (Published lexibyte on PyPI), custom PyTorch Transformer architectures, and FlashAttention optimizations.
• ⚡ Asynchronous ML Systems & Backend: High-throughput REST APIs, WebSockets, multi-stage Docker environments, and job queues built with FastAPI, Celery, Redis, and Next.js.
• 📊 Applied Machine Learning & Data Pipelines: Feature engineering, DuckDB/BigQuery SQL queries over multi-million row datasets, probability calibration, and multi-modal architectures.

Zero-framework RAG Agent & MLOps architecture built from scratch for low-resource cloud deployments.

• Engineered a custom ReAct (Reason + Act) loop paired with Google's Gemini API for autonomous tool selection and grounded answers.
• Bypassed PyTorch bloat using quantized INT8 ONNX Runtime and FAISS for local, low-latency CPU vector embeddings.
• Optimized for strict sub-512MB RAM constraints via page-by-page PDF streaming, micro-batched indexing ($N=2$), and arena memory management (enable_cpu_mem_arena = False).
• Designed a Zero-Downtime MLOps pipeline using GitHub Webhooks for atomic memory-pointer swapping during live model hot-updates.

Production-grade, algorithmically optimized Byte-Pair Encoding (BPE) tokenizer published on PyPI.

• Engineered GPT-4 style regex pre-splitting (\p{L}, \p{N}) to isolate words, numbers, and punctuation.
• Built a Sennrich Frequency Dictionary during training to reduce merge checks and integrated an O(1) memoization cache during inference.
• Features base UTF-8 fallbacks to prevent Out-Of-Vocabulary (OOV) errors.
• pip install lexibyte

An end-to-end, hardware-optimized Generative AI ecosystem. Evolved from raw PyTorch math into a fully deployed Full-Stack LLM application.

• 🧠 NanoTransformer (The Foundation)

  • Engineered a GPT-2 style Transformer decoder from scratch using native PyTorch primitives.
  • Integrated FlashAttention, bfloat16 Mixed-Precision, and a custom LexiByte BPE tokenizer to achieve massive training throughput on consumer GPUs.

• 🚀 Forge-LM (The Production Evolution)

  • Scaled the NanoTransformer architecture to ~28M parameters and trained on the TinyStories dataset, utilizing * Gradient Accumulation* to bypass 6GB VRAM hardware limits.
  • Decoupled PyTorch from inference by exporting to ONNX with INT8 dynamic quantization.
  • Engineered a lightweight FastAPI + NumPy backend and stateless Docker deployment to successfully serve the AI on strict <512MB RAM cloud environments.

Automated URL threat evaluation system built on a layered multi-modal architecture.

• Combines structured URL features (ISCX) and linguistic signals (PhiUSIIL) via a Soft-Voting Fusion aggregator.
• Wrapped XGBoost with Platt Scaling (CalibratedClassifierCV) for accurate probability output.
• Deployed with FastAPI, Streamlit, and orchestrated via Docker Compose.

Custom zero-dependency Neural Network Engine with real-time web visualization.

• Built Dense layers, ReLU, and SoftmaxCrossEntropy loss purely in raw NumPy matrix calculus.
• Integrated FastAPI WebSockets to stream real-time training metrics (Epoch/Loss/Accuracy) directly to an interactive React + HTML5 Canvas client interface.

• AI / ML Intern @ FlyRank AI (Jul 2026 – Present): Building CTR Opportunity Scoring decision-support ML models across 70M+ row production datasets using DuckDB and BigQuery (Self-paced role). Earned Anthropic Academy certifications for Claude/Claude Code.
• Data Science & ML Apprentice @ Skill Shikshya (Apr 2026 – Jul 2026): Completed rigorous hands-on engineering track spanning scratch-pad ML math, deep learning architectures, vector computation, and production MLOps. Successfully defended final project in July 2026 (certification pending).
• Fullstack Engineer Intern (Academic Requirement) @ Walkers Hive IT Professionals (Oct 2025 – Dec 2025): Completed mandatory degree internship by leading the independent architecture of AcademiaOS MVP (FastAPI, Next.js, Celery, Docker, HTTP-only cookie auth, RBAC).
• Software Engineer (Contract) @ Nextwave Technology (Apr 2025 – Jul 2025): Led Play Store launch for the " Academia" mobile app, fixed legacy Flutter codebases, and migrated corporate sites to dynamic Next.js platforms.
• Software Engineer (Contract) @ Walkers Hive IT Professionals (Nov 2024 – Apr 2025): Built e-commerce admin panel with React/MUI/Redux-Saga and deployed full-stack Next.js frontends over legacy PHP backends.
• Android Development Intern @ CodSoft (Dec 2023 – Jan 2024): Developed Flutter applications with Firebase authentication, local persistence, and BLoC state management.

• Bachelor of Computer Application (BCA) (2021 – 2026)
  Niharika College of Management and Information Technology (Tribhuvan University, Nepal)
  Status: Coursework completed; final examinations starting August 11, 2026 (degree completion mid-August 2026).

📫 Let's Connect:

• Email: imbibek8366@gmail.com
• LinkedIn: https://www.linkedin.com/in/bibek-dhakal-771ba5334/