Skip to content

// Open to internships & research collaborations

Pratik Sharma

_

I build retrieval-augmented and applied-ML systems, then ship them as real, usable software. Most of my time goes into the gap between a model that works in a notebook and a system that holds up in production — evaluation, failure analysis, and the backend plumbing around it.

Nepal
PS

research focus

Retrieval-Augmented Generation

queryretrieverankgenerateanswerfailure-mode check

// about

Applied AI, end to end — not just the model.

I'm a Computer Science undergraduate based in Nepal, focused on the applied side of AI: retrieval-augmented generation, NLP, and the infrastructure that makes machine learning systems trustworthy enough to actually deploy.

My coursework and side projects sit at the intersection of three things — large language models, the data engineering that feeds them, and the backend work needed to put them in front of real users. I'm equally comfortable profiling a Spark job and shipping a Next.js front end for it.

Right now I'm most interested in why RAG pipelines fail silently — retrieval misses, stale context, hallucinated citations — and how to detect those failures before a user sees them, particularly in medical and other high-stakes domains.

goals

Looking for internship, research-assistant, or junior engineering roles where I can work on applied ML systems end-to-end — from data pipeline to model to the product surface people touch.

currently

Higher Secondary Education

Global School of Science

20212022

strengths

  • Translates research papers into working pipelines
  • Comfortable across the stack: data, model, API, UI
  • Writes for failure cases first, not just the happy path
  • Self-directed — most projects below started outside any classroom

// skills

A stack built for retrieval-heavy AI systems.

Grouped by where the work actually happens: languages I write daily, the web layer I ship in, the ML/AI core, the data underneath it, and the ops that keep it running.

What I reach for day to day

Python90%
JavaScript40%
TypeScript30%
Java10%
SQL70%
C++70%
C64%

// projects

Things I've built.

From a research-grade RAG diagnostic system to a full-stack marketplace — each one chosen because it forced a different kind of hard problem.

proj/clinicalResearch

Clinical RAG Failure Classification System

Catching retrieval failures before a clinician sees them

A diagnostic layer that sits inside a medical retrieval-augmented generation pipeline and classifies why a given answer might be untrustworthy — missing context, irrelevant retrieval, or contradicted-by-source generation — instead of only scoring the final answer.

QueryRetrieveClassify failure modeFlag or answer

features

  • · Medical document retrieval over a clinical corpus
  • · A dedicated retrieval-failure detection model, separate from the generator
  • · End-to-end RAG pipeline with swappable retriever/generator components
  • · LLM integration for answer synthesis with failure-aware fallbacks
PythonPyTorchLangChainFAISSscikit-learn

Challenge — Most RAG evaluation only scores the final answer, which hides whether a wrong answer came from bad retrieval or bad generation. Building a labeled failure-mode dataset and a classifier that separates the two was the core technical problem.

Impact — Gives a concrete, inspectable reason when a clinical RAG answer should be trusted less — a step toward RAG systems that fail loudly instead of silently in high-stakes domains.

Code Demo coming soon
proj/grubmateIn progress

GrubMate

A full-stack food delivery platform, built end to end

A multi-sided food delivery platform covering customers, restaurants, and delivery riders — order placement, restaurant-side order management, and live delivery tracking in one system.

features

  • · User management and authentication across customer and restaurant accounts
  • · Restaurant management dashboard for menus and incoming orders
  • · Live delivery tracking
  • · Order management from placement through fulfillment
DjangoPostgreSQLReactREST APIs

Challenge — Coordinating state across three different user roles (customer, restaurant, rider) in real time without the order pipeline getting out of sync was the main design challenge.

Impact — A complete, working reference for multi-sided marketplace architecture — the kind of system design question that shows up in full-stack interviews.

Code Demo coming soon

// research

Where I want to spend the next few years.

Centered on making retrieval-based AI systems reliable enough to trust in domains where a wrong answer actually costs something.

01

Retrieval-Augmented Generation

Failure detection, evaluation, and reliability of RAG pipelines beyond top-line accuracy.

02

Medical AI

Applying LLMs and retrieval systems to clinical text where errors carry real cost.

03

Explainable AI

Making model decisions — especially retrieval decisions — inspectable rather than opaque.

04

Natural Language Processing

Core NLP methods underpinning retrieval and generation systems.

05

Knowledge Graphs

Structured knowledge as a complement to vector retrieval for grounded generation.

// education

Academic background.

Higher Secondary Education

Global School of Science · Baneshwor,Nepal

20212022

relevant coursework

PhysicsChemistryMathematicsComputer ScienceEnglish

Bachelor in Computer Science

Khwopa Engineering College · Bhaktapur, Nepal

2023...

relevant coursework

Machine LearningArtificial IntelligenceComputer NetworksDatabase SystemsComputer Organization and Architecture

// open source

What I've been shipping on GitHub.

@PratikSharma264
Pratik Sharma's GitHub contribution graph
Pratik Sharma's GitHub stats
Pratik Sharma's most-used languages

// contact

Let's talk about an internship, role, or project.

Whether it's an internship, a research collaboration, or a freelance build — I read every message myself.

sharmapratikabcd@gmail.com+977 9841327376
Nepal

Open to internships & research collaborations