Hi, I'm Prashant.
AI Engineer
Building intelligent systems — LLMs, RAG pipelines,
and autonomous agents that solve real-world problems.
Profile
Name Prashant Raj Bista
Location Kathmandu, Nepal
Email prashant.bista.18@gmail.com
LinkedIn prashant-raj-bista
GitHub @prashantrajbista
Status Open to opportunities
My Story
Building the Future with AI
I'm an AI Engineer currently working full-time at IdeaBoxAI and Azminds Services in Kathmandu. I hold a BE in Electronics, Information & Communication Engineering from Tribhuvan University, IOE, Thapathali Campus.
I specialise in building production-grade AI systems — RAG pipelines, LLM-powered APIs, and generative AI products. My work sits at the intersection of backend engineering and applied AI, turning complex LLM capabilities into reliable, scalable software that solves real-world problems.
Technical Skills
Technologies I use to build AI-powered systems
AI & LLMs
ML & Data
Backend & APIs
Vector DBs & Search
Infra & Tools
Electronics & Embedded
Featured Projects
AI systems and engineering work I'm proud of
Experience
My journey in engineering and AI
Junior AI Engineer · IdeaBoxAI
Building production AI systems across the full ML spectrum at IdeaBoxAI. Work spans developing time series forecasting models, classification and regression pipelines, and generative AI products — all served through FastAPI microservices backed by vector databases for semantic retrieval. Deeply embedded in the Anthropic ecosystem, using the Claude API and Claude SDK to build tool-use agents, structured-output pipelines, and context-aware AI features shipped to end users.
Associate Software Engineer · Azminds Services
Working on-site in Kathmandu designing and deploying RAG pipelines and LLM-integrated backend systems for enterprise clients. Responsibilities span building retrieval-augmented generation architectures, fine-tuning and evaluating large language models, developing REST APIs to serve AI features, and ensuring system reliability and performance in production environments.
BE in Electronics, Information & Communication Engineering
Tribhuvan University, IOE, Thapathali Campus · Kathmandu, Nepal. Four-year programme covering signal processing, computer networks, embedded systems, and software engineering — with a final-year focus on applied AI and machine learning projects.
Research
Academic projects and research publications
Nepali-to-English Speech Translation with Prosody Preservation
Tribhuvan University, IOE, Thapathali Campus — BE Major Project Final Report
An end-to-end speech-to-speech translation pipeline that converts Nepali spoken audio to natural-sounding English speech while preserving prosodic features — tone, stress, and rhythm. The system chains three state-of-the-art models: wav2vec 2.0 for Nepali ASR (fine-tuned on SLR143, SLR43 and Mozilla Common Voice), mBART for Nepali-to-English neural machine translation (fine-tuned on a manually curated parallel corpus), and F5-TTS for prosody-aware English speech synthesis. A custom parallel Nepali–English audio dataset was created to train and evaluate prosody transfer.
Deep Generative Modeling for Automated Image Manipulation via Text-Guided Prompts
NCE Research Grant — IOE, Thapathali Campus
Proposed a novel text-driven image editing framework built on Stable Diffusion that addresses two critical gaps in existing methods: simultaneous multi-object editing within a single prompt, and a undo/redo mechanism that removes edits without residual artifacts. Text prompts are parsed with NLP models, segmented into per-object instructions, and applied iteratively through a latent-diffusion pipeline. Evaluated using LPIPS and MS-SSIM metrics on COCO and ImageNet datasets.
Role: Mathematical Modeling for Simultaneous Edits (Linear Algebra, Probability, NLP)
Vision-aided Mechanical Design for an Autonomous Rubik's Cube Solver
Tribhuvan University, IOE, Thapathali Campus — BE Minor Project
Built an end-to-end mechatronic system that autonomously solves a scrambled 3×3×3 Rubik's Cube. A webcam feeds raw frames into a YOLOv8 colour-detection model that maps cube faces to a state string; Kociemba's Algorithm computes the optimal move sequence (≤20 moves); an Arduino UNO drives three stepper-motor axes to physically execute each rotation. A companion C# GUI virtualises the cube state in real time.
Recent Developments in SLAM and Drone Autonomy: A Five-Year Perspective
Independent Review — Preprint, August 2025
A survey synthesising five years (2020–2025) of progress in Simultaneous Localisation and Mapping (SLAM) and UAV autonomy. Covers resource-constrained SLAM for nano-UAVs (NanoSLAM), dynamic environment adaptations (DynaVINS, DynaGSLAM), depth-image quality improvements for 3D reconstruction, AI-driven drone control via foundation models and LLM-generated command systems, and drift-free localisation through digital twin alignment.
Let's Talk
Freelance projects · Full-time AI roles · Research collaborations
AI Engineer building intelligent systems with LLMs, RAG, and agents. Based in Kathmandu, Nepal. Available for global remote work.
Email