Skip to main content
Tilak Parajuli

Tilak Parajuli

AI/ML Engineer

Email GitHub LinkedIn Scholar ORCID

I design production AI systems for decisions that have to be right: autonomous LLM pipelines and the evaluation that keeps them honest.

Download CV (PDF) Get in touch

I work where applied LLMs meet real consequences. My focus is multi-agent pipelines, retrieval-augmented generation, and fine-tuned models, deployed on AWS. As a Data Scientist at Climate Clean Solutions, I build autonomous systems that contest property-tax valuations end-to-end, enforcing measurable reduction thresholds across thousands of parcels.

Previously, as an AI/ML Research Fellow at Fusemachines, I shipped real-time gaze detection into a live proctoring product and built the model-evaluation tooling teams relied on across their forecasting systems. Earlier, at NAAMII, I worked on computational biology for radiation oncology under Dr. Taman Upadhaya (Cedars-Sinai).

B.Sc. in Computer Science, Tribhuvan University.

Selected Work

Multi-Agent Tax Valuation Pipeline

Climate Clean Solutions, 2026

Autonomous LLM system for property tax appeals. RAG-driven comparable retrieval, automated valuation modeling, and QA enforcement across thousands of parcels.

RAG Pipeline for Job Retrieval

Python, LlamaIndex, ChromaDB, Cohere, Docker

Hybrid retrieval combining BM25 sparse search with dense vector embeddings and Cohere reranking, served via FastAPI.

Travya: Multi-Agent Travel Platform

LangGraph, React, PostgreSQL, Redis, Docker

Specialized research, planning, and booking agents with streaming AI and third-party integrations.

Text Summarization (LSA + T5)

Python, Transformers, NLP

Dual extractive/abstractive summarizer. Fine-tuned T5 alongside LSA, evaluated with ROUGE metrics.

Recent