model loaded · ankit-rai · Kathmandu, Nepal

ANKIT RAI

> AI engineer shipping

AI engineer building production RAG systems, autonomous agents and fine-tuned LLMs. I turn research-grade models into deployed, containerized products.

view inference log ↓POST /contact

ARCHITECTURE

RAG · agents · LoRA

TRAINING DATA

B.Sc. CSIT · Tribhuvan Univ.

DEPLOYED AT

Frost Digital Ventures

STATUS

accepting new prompts

[01] — who am I

// System Prompt

system_prompt.yaml

role:"Associate AI Engineer @ Frost Digital Ventures"

obsession:"making LLMs actually useful in production"

specialties:["RAG", "Graph RAG", "AI agents", "LoRA/PEFT"]

serves_with:["FastAPI", "Docker", "Kubernetes"]

base:"Kathmandu, Nepal 🇳🇵"

temperature:0.7 # creative, but grounded

I'm Ankit Rai — an AI engineer who lives in the gap between research papers and running products.

My day job is building knowledge-grounded chatbots on custom FAISS vector stores, pushing retrieval further with Graph RAG, wiring up autonomous agents, and fine-tuning open-source LLMs — then benchmarking them honestly against the closed-source giants.

Everything ships in Docker behind FastAPI, because a model that isn't deployed is just a very expensive matrix.

[02] — work experience

// Fine-Tuning

epoch 03 · Oct 2024 — Presenttraining in progress

Associate AI Engineer · Frost Digital Ventures

Designed and deployed domain-specific chatbots on RAG pipelines with custom FAISS vector databases for knowledge-grounded conversations.
Implemented Graph RAG retrieval for relational understanding and multi-document reasoning.
Built autonomous AI agents and augmentation utilities for task-driven reasoning workflows.
Fine-tuned open-source LLMs (Hugging Face) and benchmarked against OpenAI & Gemini on accuracy, latency and cost.
Containerized and served AI apps with Docker + FastAPI for scalable cloud deployment.

epoch 02 · Aug 2024 — Oct 2024

AI Intern · Frost Digital Ventures

Preprocessed and visualized large datasets; engineered features; maintained reproducible training notebooks.
Trained CNN and RNN architectures (incl. transfer learning) for object detection and classification.
Implemented NLP pipelines and experimented with GANs and LLM applications.

epoch 01 · May 2024 — Jul 2024

Python / Django Intern · Orbitech Nepal

Built Django apps with DRF REST APIs, authentication and PostgreSQL/SQLite.
Implemented web-scraping pipelines (BeautifulSoup) for structured data extraction.
Used Git/GitHub and deployed projects to test environments for QA review.

[03] — selected projects

// Inference Log

out[0] · end-to-end RAG

Simple-RAG-Chatbot

Retrieval-augmented generation pipeline pairing FAISS vector search with open-source LLMs for context-aware Q&A over PDFs, with CUDA-accelerated embeddings.

FAISSHugging FaceGemma-2-2b-itCUDA

out[1] · Nepali OCR

CharacterRecognizer

OCR pipeline classifying handwritten Nepali characters with CNN/ResNet, image preprocessing and fine-tuned pretrained models on custom datasets.

CNNResNetTransfer Learning

out[2] · seq. classification

NewsClassification

Sequence models categorizing news into topics; compared RNN/LSTM/GRU variants on accuracy vs efficiency with F1-driven evaluation.

RNNLSTMGRUTensorFlowPyTorch

out[3] · -90% params

Sentiment Analysis · LoRA

Fine-tuned DistilBERT with Low-Rank Adaptation — 90% fewer trainable parameters at equal accuracy on IMDb / SST-2.

TransformersLoRAPEFTPyTorch

> full history: github.com/2001-Ankit

[04] — skills & stack

// Benchmarks

PROGRAMMING

suite_0

PythonJavaScriptC#SQL

ML & LIBRARIES

suite_1

PyTorchTensorFlowHugging Facescikit-learnFAISSPEFT / LoRA

TOOLS & DEV

suite_2

GitDockerKubernetesFastAPIStreamlitPostgreSQLNeo4jLinux

CLOUD

suite_3

GCPAWSAzure

[05] — education

// Training Data

2019 — 2023

B.Sc. Computer Science & IT (CSIT)

Tribhuvan University — Orchid International College

Aggregate 75.57% · Linear Algebra, Probability & Statistics, DSA, Machine Learning, DBMS, OS, Networks, Cloud Computing

2017 — 2019

Higher Secondary (+2), Science

Reliance International Academy

Chabahil, Kathmandu

[06] — get in touch

// API Reference

# The only endpoint that matters

POST /ankit/collaborate

{
  "project": "something ambitious",
  "involves": ["RAG", "agents", "LLMs", "or anything AI"],
  "timeline": "yesterday, preferably"
}

200 OK— response time: usually < 24h

Have a problem that needs an AI engineer who ships? My inbox is an open port.

itsmeankit101@gmail.com (+977) 9813928960

GitHub LinkedIn