Nishant ParajuliLet’s talk
काठमाडौँKathmandu, Nepal27.7172° N · 85.3240° E · 1,400 m
Nepal Time · UTC+5:45

AI, built to run.

I’m Nishant. I build AI applications, engineer their backends, and operate the infrastructure behind them.

Explore the workAvailable now · Full-timeNepal / Remote
  • AI systems
  • Backend engineering
  • Personally operated infrastructure

Running live: gradient descent on a drifting loss surface. Your cursor reshapes it.

01 Selected work

Less talk. More built.

A selection of systems where AI meets the messy, interesting work of making things useful.

01 / 04Retrieval systems / Full-stack2026

Spoiler-aware webnovel wiki

Useful answers. No reading ahead. A retrieval system built around the reader’s place in the story.

  • FastAPI
  • React
  • PostgreSQL
  • pgvector
  • LLMs
Read the case study
A boundary for knowledgeInteractive concept
What do we know so far?
Chapter 12 of 25
CH. 04

An unfamiliar arrival

Character introduced
CH. 10

A name in the archives

Entity relationship
CH. 16

Beyond your chapter

Excluded from context
CH. 22

Beyond your chapter

Excluded from context
Only the story you’ve read.2 of 4 example passages available to retrieval

Move the boundary. Watch the available context change. Illustrative chapter data.

02 / 04Speech synthesis / Language engineeringIndustry work

Nepali text-to-speech

From written Nepali to spoken sound: custom pronunciation handling, VITS2, and a RingFormer vocoder.

  • Python
  • G2P
  • VITS2
  • RingFormer
  • Speech synthesis
Read the case study
Written → spokenPronunciation explorer
देश
Written Nepali · inherent vowel dropped
FINAL शOmit the inherent vowel.

The final consonant is spoken without its inherent vowel.

  1. 01Nepali G2PPronunciation
  2. 02VITS2Speech & timing
  3. 03RingFormerWaveform decoder

Linguistic examples and a schematic waveform. No generated audio.

03 / 04Generative AI / Backend architectureIndustry work

Multimodal book generation

Turning a book brief into a structured, illustrated draft through a staged authoring workflow.

  • Python
  • FastAPI
  • LLMs
  • Multimodal AI
  • Background jobs
Read the case study
From brief to bookWorkflow explorer
01 · PlanA shared outline
  1. Audience & learning goals
  2. Chapter progression
  3. Section structure
01

Give every chapter a purpose.

Research and an editable outline establish the book’s direction before content generation.

An illustration of the authoring flow, using sample page layouts.

04 / 04Applied machine learning / Backend2024

Vendor-centric ecommerce

Bringing prediction, recommendations, and customer insight into the everyday flow of a vendor’s business.

  • Python
  • Django
  • Machine learning
  • Analytics
Read the case study
Intelligence for the vendorFeature explorer
InputSales history
ML + Django
The application

Plan for what comes next.

Future-sales prediction adds a forward-looking signal to the vendor workflow.

A diagram of the feature flow. Illustrative inputs.

  • Retrieval
  • Speech
  • Computer vision
  • Multimodal generation
  • Backend engineering
  • Infrastructure

02 Experience

Where I’ve been building.

Models, language systems, and the infrastructure around them. Open a role to explore the work.

AI EngineerInnovate Nepal Group

Jun 2024 — Sep 2026KathmanduView 7 contributionsHide contributions

  • Computer vision

    Designed and deployed a scalable facial recognition system using computer vision with 98.5% accuracy across diverse demographics.

  • Classroom speech intelligence

    Developed a real-time classroom transcription and intelligence pipeline using NVIDIA Nemotron, with speaker diarization, sentiment analysis, lesson summaries, and interactiveness measurement.

  • Contextual retrieval

    Engineered a multidisciplinary AI teaching assistant using LangChain and contextual RAG with parent-child chunking to automate lesson planning and student curricula across multiple schools.

  • Automated assessment

    Engineered an automated assessment system leveraging NLP for real-time exam feedback, reducing grading workload while maintaining scoring consistency.

  • Nepali speech synthesis

    Developed a custom Nepali G2P frontend for schwa deletion and exceptions, and trained a VITS2 text-to-speech model with a RingFormer vocoder.

    Explore the project

  • Multimodal book generation

    Architected a multimodal personalized book generator using LLMs and image-generation models, with a pre-generated structural skeleton for narrative and visual consistency.

    Explore the project

  • Cloud infrastructure & delivery

    Orchestrated scalable AWS and GCP infrastructure for containerized FastAPI microservices using Terraform, Kubernetes, Docker, and GitLab CI.

AI/ML DeveloperIcebrkr

Mar 2024 — Jun 2024KathmanduView 3 contributionsHide contributions

  • Classification & language processing

    Optimized lightweight edge models for text classification, enhanced with rule-based approaches and Named Entity Recognition.

  • Datasets & recommendations

    Created datasets and implemented a recommendation system using word embeddings and ranking algorithms.

  • Computer vision

    Developed computer vision techniques for facial recognition on low-end devices and deployed them via FastAPI.

AI InternInfoDevelopers Pvt Ltd

Nov 2023 — Feb 2024KathmanduView 2 contributionsHide contributions

  • Machine learning in applications

    Implemented machine learning models inside Django projects, focusing on performance optimization.

  • Image classification & detection

    Used OpenCV and PyTorch for computer vision tasks including image classification and object detection.

03 Infrastructure

Built by me. Run by me.

The work doesn’t end at git push. A look inside the infrastructure I operate.

Architecture view · live metrics unavailable
From a question to an answerPortfolio assistant · illustrative flow
02 / 04

The place where the pieces meet.

FastAPI serves the portfolio, handles contact submissions, and prepares context for the AI assistant. Request middleware records timings, while health endpoints expose service and database checks.

Host
Personal server
Edge
Cloudflare tunnel
Runtime
Docker
Data
PostgreSQL + pgvector
Personal server. Real responsibilities.Inside the infrastructure

04 About

The whole way through.

My work lives between an idea that’s promising and a system that actually runs.

I’m an AI engineer based in Kathmandu, Nepal. I earned my BSc (Hons) in Computing at Islington College in 2024. I work across retrieval, computer vision, speech, and the backend services that turn models into applications.

That curiosity extends beyond the code. I manage my personal VPS, so deployment and operation are part of the work, too.

Download my CV PDF

CV updated

Connect on LinkedIn
  1. 01

    Intelligence

    Contextual RAG, computer vision, NLP, speech, and multimodal generation.

  2. 02

    Engineering

    Python, FastAPI, Django, PostgreSQL, and services built around the problem.

  3. 03

    Operation

    Linux, Docker, cloud infrastructure, and a personal VPS I run myself.

Tools and technologies

  • Python
  • Java
  • SQL (PostgreSQL, MySQL)
  • JavaScript
  • HTML/CSS
  • PyTorch
  • Django
  • FastAPI
  • Scikit-learn
  • OpenCV
  • Pandas
  • NumPy
  • Matplotlib
  • PySpark
  • Linux
  • Docker
  • AWS
  • GCP
  • Terraform
  • Kubernetes
  • Apache Airflow
  • Git
  • GitLab CI
  • Data Structures & Algorithms
  • Machine Learning
  • Computer Vision
  • NLP
  • RAG Systems

The lab

Still curious?

Ask about the work or explore the API that powers this portfolio.

Step into the lab

05 Contact

Let’s build something that works.

Have an AI problem, a backend challenge, or a team I should meet? I’d like to hear about it.

nishant.parajuli123@gmail.com
Based in Kathmandu. Open to a good conversation.

Available now · Full-time · Nepal / Remote

Send Nishant a message

Prefer your own inbox? The email link works too.