B.Sc. Computer Science and Information Technology
Soch College of IT, Pokhara
Final semester (8th), GPA 3.4/4.0, Batch Topper (2nd and 3rd semesters)
Jyoti Pokhrel / AI Engineer, Researcher
Applied AI researcher who turns open questions into working systems: agentic AI, RAG, LLM evals and multimodal models.
About
AI engineer and applied researcher in LLM agents, RAG, knowledge graphs and multimodal models, with a focus on evaluation pipelines and harness engineering.
At Ninebar I build agent prototypes, retrieval benchmarks and LLM-as-judge evals. Before that I shipped a computer vision model used by 3,000 people monthly. Published in INJET 2025.
Education
Microdegree in AI
Fusemachines 2025
Nepal AI School, NAAMII, Dec 2025
Experience
Applied AI Research Intern
Ninebar
Singapore, Remote
- Delivered 5 AI proofs of concept in 6 weeks across LLM agents, knowledge graphs and document AI.
- Built a text-to-Cypher LLM agent over a Neo4j knowledge graph, with schema-validated queries.
- Built an LLM-as-judge evaluation harness (DeepEval, G-Eval) benchmarking 23 model configurations.
- Benchmarked 4 graph and RDF stores for LLM retrieval, with 0 data loss on round trip.
- Built multimodal chart extraction with vision-language models.
AI Engineer Trainee
SkinPal AI
Nepal, Remote
- Shipped a skin analysis computer vision model (acne severity, pores, blackheads, whiteheads) for 3,000 monthly users at under 10 ms per photo; tested on 2,500+ photos.
- Cut false positives 30% by adding lighting normalisation with OpenCV before inference.
- Automated an LLM TikTok pipeline: 30 scripts a week, script writing cut from 2 hours to 10 minutes.
- Built a Reddit AI agent that scans 20 subreddits and drafts 40 contextual replies a day.
AI Fellow
Fusemachines
Nepal, Remote
- Completed 10+ end-to-end ML projects across computer vision, generative modeling and signal processing.
- Tracked every project in MLflow, from experiment runs to deployment-ready models.
Selected Work
Each one can be inspected: a repository or a measured result.
01 Global IME AI/ML Hackathon 2026 · 2nd Runner-Up
Automated Lending Lifecycle, Multi-Agent System
Built a 5-agent LangGraph pipeline for loan decisions over 100K applicants and 4M+ records. Reached R² 0.938 on income, 0.965 on credit score and 0.803 AUC on default risk.
LangGraphMulti-agentXGBoostSHAP
02 Repo ↗
Repo Explainer
Hybrid RAG for code Q&A: BM25 and BGE retrieve top-20 each, fused to top-5 with Reciprocal Rank Fusion. AST-aware chunking with tree-sitter for 7 languages; deployed with FastAPI, Pinecone and Render.
RAGBM25BGETree-sitterFastAPIPinecone
03 Repo ↗
Literature Assistant
Research assistant over 4 paper sources with UMAP + HDBSCAN clustering, LLM research-gap reports and a 4-stage CI.
UMAPHDBSCANLLMD3.jsFastAPI
Research & Writing
Skills
01
LLMs & Agents
- LangGraph
- LangChain
- LlamaIndex
- RAG
- Hybrid retrieval
- Embeddings
- Vision LLMs
- Prompt engineering
02
Evaluation
- Evaluation pipelines
- Harness engineering
- LLM-as-judge
- DeepEval
- Benchmarking
- Reproducibility
- MLflow
03
ML & Vision
- PyTorch
- scikit-learn
- XGBoost
- SHAP
- OpenCV
- UMAP
- HDBSCAN
04
Backend & Data
- FastAPI
- Docker
- PostgreSQL
- pgvector
- Pinecone
- MongoDB
- Cypher
05
Languages & Tools
- Python
- SQL
- Git
- Linux
- GitHub Actions
- AWS
- GCP
Recognition
- 01 Joint Secretary, CSIT Association of Nepal, Pokhara (Mar 2025 - Jul 2026): ran tech workshops
- 02 Harvard Aspire Leaders Program
- 03 Women in Data 2025 Scholar
- 04 IDEATHON FemTech, 1st Runner-Up
Contact
Open to AI engineering and research roles.
Email is the best place to start.
[email protected]