← Back to portfolioA4 / Letter — margins: default

Praansu Karmacharya

Junior AI Developer — ML models, RAG systems, agent loops.

Kathmandu, Nepal · +977 9818759455 · Praansu12@gmail.com · praansu.github.io · github.com/Praansu

Summary

Junior AI Developer intern at Aviyaan Tech (Mar 2026 — Present), building and experimenting with ML models — ensemble methods for traffic-count estimation and road-damage segmentation and classification. BSc (Hons) Computing with AI at Islington College, Kathmandu (2025 — 2028). Self-directed builds in RAG pipelines, agent tool-calling, and vision models, each with honest evaluation.

Experience

Junior AI Developer (Intern) — Aviyaan Tech

Mar 2026 — Present · Kathmandu
  • Build and experiment with ML models for vision and counting tasks, from data prep through training and evaluation.
  • Ensemble methods for traffic-count estimation under varied road conditions.
  • Road-damage segmentation and classification experiments on pavement imagery.
  • Report per-class metrics and failure cases instead of headline accuracy, and iterate from there.

Selected projects

  • AI Research Agent — no-framework agent loop over private documents plus web search, streaming every tool call live to the browser.Python · FastAPI · ChromaDB · SSE · github.com/Praansu/ai-research-agent
  • Small-Agent Reliability Study — 9 open-weight models (1B–9B) scored as tool-using agents across 31 capability and 14 reliability tasks; Qwen 2.5 Coder 7B led at 85% composite reliability.Python · Ollama · pandas · LaTeX · github.com/Praansu/small-agent-reliability
  • Vehicle Image Classifier — ResNet-18 transfer learning, 4 classes, training images labeled with a self-built Tkinter annotation tool; confusion matrix and per-class accuracy; served via FastAPI, containerised.PyTorch · FastAPI · Docker · github.com/Praansu/vehicle-image-classifier
  • EcoVerda (demo storefront) — full-stack e-commerce demo: cart persistence, credentials auth, orders and reviews.Next.js 16 · TypeScript · Prisma · Stripe · praansu.github.io/eco-verda

Education

BSc (Hons) Computing with AI — Islington College, Kathmandu

2025 — 2028

Focus on machine learning, AI systems, and full-stack development.

+2 Science (Computer Science) — Global School of Science

2024 — 2025

Higher secondary education in the Science stream with Computer Science.

Skills

ML / AI
PyTorch, scikit-learn, XGBoost, OpenCV, sentence-transformers, pandas
LLM
FastAPI, ChromaDB, RAG pipelines, agent tool-calling, Groq / Ollama
Full-stack
Next.js, TypeScript, React, Tailwind, Prisma, PostgreSQL / SQLite
Ops
Docker, Git / GitHub Actions, Linux / shell, LaTeX

Tip: enable “Background graphics” for the full blueprint look.