Available for AI Engineer & FDE roles

Hi, I'm Asmi Building AI Systems That Ship

AI Engineer passionate about retrieval, agents and data systems that answer real questions, and about measuring whether they are right.

Currently building: PolicyGuard v2, adding an agent layer, an evaluation set and CI to my offline RAG compliance suite.
AI Engineering,
RAG & Agent Systems,
Data Pipelines &
Forward Deployed Work.
Portrait of Asmi Basnet

11

WebMCP tools in Nepal Disaster Watch

34k+

Synthetic EHR Records Queried

77

Districts cross-referenced by one tool

PythonSQLFAISSPyTorchDuckDBDockerStreamlitGo

Things I've Built

End-to-end AI and data systems built around retrieval, evaluation and clear architecture, from RAG and agents to warehouses and ML.

Selected Work

Click a project to feature it above

Case Study

Nepal Disaster Watch: turning scattered disaster evidence into one view that both a responder and an AI agent can use.

The problem

After floods and landslides cut off villages in Nepal, the evidence a responder needs is spread across separate sources: NDRRMA/BIPAD incidents, DHM river gauges, Department of Roads closures, UN GDACS alerts, Copernicus GloFAS forecasts and Copernicus EMS damage grading. A responder has to check each one separately.

What I built

A live Leaflet map app with 11 native WebMCP tools, built with Sushrit Kafle for OpenAI's WebMCP Challenge.

  • 6 read tools, including cross_reference_district, which compares evidence from every source across all 77 districts.
  • 5 write tools that pan the map, filter records and open drill-downs in sync with what the human is looking at.

What it measures and shows

11WebMCP tools (6 read, 5 write)
6institutional data sources
77districts cross-referenced
3feed states: live, unreachable, snapshot

Every feed reports its own status, so a source that is down shows up as unreachable instead of failing silently. That matters when someone is making decisions from the screen.

Honest limits

This is an independent, open-source project and not an official warning service. I am not quoting accuracy or speed numbers here. The counts above describe what it does.


How I Build?

01

Questions.

Problem framing. Data checks. What "correct" means.
SchemasRelational design
Data QualityValidation rules
02

Retrieval.

Embeddings. Search. Baselines before complexity.
FAISSVector search
PyTorchModels
03

Engines.

Columnar queries. Clean transforms. Tested pipelines.
DuckDBFast analytics
SQL & GoTransforms and validators
04

Explainability.

Show why an output was produced so people can trust it.
SHAPExplanations
XGBoostRisk models
05

Products.

Package it. Document it. Put it in front of users.
DockerPackaging
StreamlitInterfaces
FastAPIServices

Experience

Hands-on work spanning AI engineering, RAG pipelines, ML models, spatial data science, and embedded robotics.

Aug 2025 – Present · Hybrid01

Research Assistant (AI & RAG Systems)

Saint Louis University · St. Louis, MO
  • Engineered an enterprise RAG (Retrieval-Augmented Generation) pipeline for automated HR compliance Q&A, leveraging FAISS for low-latency semantic vector retrieval across institutional policy documents.
  • Integrated local, quantized LLM inference engines to ensure offline execution, complete data isolation, and strict compliance with confidential institutional privacy standards.
  • Established multi-stage guardrail architectures using dynamic cosine similarity thresholds, curtailing hallucinations and out-of-domain queries by over 30%.
  • Implemented automated citation validation algorithms to map generated answers directly to underlying policy documents, guaranteeing fully auditable, verifiable outputs.
  • Built structured JSONL audit logging pipelines and observability monitors to track retrieval relevance, citation coverage, and latency metrics across complex edge-case evaluations.
PythonFAISSQuantized LLMsOffline RAGGuardrailsAudit Logging
Jul 2023 – May 2024 · Remote02

Data Scientist

Rabin Nirman Sewa · Kathmandu, Nepal
  • Analyzed spatial, geographic, and field-survey datasets using Python and SQL to optimize transmission tower placement and right-of-way (ROW) clearance along complex terrain.
  • Developed automated tracking pipelines and interactive operational dashboards (Power BI / Excel) to monitor material procurement, line stringing progress, and labor allocation across high-voltage distribution sites.
  • Performed exploratory data analysis (EDA) on contractor delivery timelines and weather disruptions, improving project schedule forecasting accuracy and resource utilization by 15%.
PythonSQLSpatial AnalysisPower BIForecastingEDA
Jun 2024 – Jul 2024 · Hybrid / Remote03

Data Visualization & Project Management Intern

Excelerate & GlobalShala · Saint Louis University
  • Collected, cleaned, and transformed raw digital marketing campaign datasets into interactive Power BI and SQL dashboards, delivering data-driven visual storytelling on engagement trends and KPIs.
  • Coordinated multi-phase technical and marketing projects at GlobalShala, defining project scopes, tracking milestones, and presenting progress reports to senior leadership.
  • Implemented automated validation and data-cleaning checks across sensitive datasets to ensure reporting accuracy and governance compliance.
Power BISQLData VisualizationProject ManagementKPI Dashboards
Dec 2022 – Jul 2023 · Hybrid04

Researcher & Data Analyst

Entegra Sources Pvt. Ltd. · Kathmandu, Nepal
  • Cleaned, transformed, and mapped survey, transactional, and operational data into standardized relational schemas using Python (Pandas, SQLAlchemy) and SQL.
  • Analyzed 5,000+ survey responses in SPSS and R (regression, ANOVA, chi-square) and authored statistical methodology and audit documentation.
PythonSQLRSPSSStatistical ModelingRelational Schemas
Aug 2018 – Oct 2022 · On-site05

Robotics Hardware Lead & Treasurer

Electronics Project Club · Kathmandu, Nepal
  • Programmed microcontroller firmware (C/C++ & Python) for autonomous and remote-controlled robotics competing in inter-college engineering competitions.
  • Designed and integrated sensor arrays (ultrasonic, IR, and IMU) to implement real-time obstacle detection, line tracking, and PID motor-speed control algorithms.
  • Validated circuit schematics and hardware logic via simulation tools before prototyping and soldering custom PCB components, debugging edge-case sensor latency for robust performance.
C/C++PythonEmbedded SystemsPID ControlSensors & RoboticsPCB Prototyping

Education & Credentials

Graduated May 202601

M.S. in Analytics

Saint Louis University · St. Louis, MO
  • Focus on machine learning, deep learning, statistical modeling, data warehousing, and LLM prompt engineering / agent architectures.
2017 – 202202

B.E. in Electronics & Communication

Tribhuvan University · Nepal
  • Foundation in probability, statistics, signal processing, embedded microcontrollers, and digital hardware logic.
Professional Accreditations03

Certifications & Datathons

Industry & Research Credentials
  • AI Engineer for Data Scientists Associate — DataCamp (PyTorch neural network architectures, custom loss functions, scikit-learn ML pipelines).
  • Building with Claude & Anthropic API — Anthropic (Prompt caching, MCP tool integration, production evaluation loops).
  • WashU I2DB Diabetes Risk Prediction Datathon — WashU Institute for Informatics & SLU (Built XGBoost + LightGBM ensemble, AUC 0.8520, Optuna, 59 clinical features, SHAP explainability).
  • Microsoft Certified: Power BI Data Analyst (PL-300) — Microsoft.

About Me

About the engineer

Hi, I'm Asmi.

I come from an electronics and communication background and hold a Master's in Analytics. I like systems that are measurable: a retrieval pipeline with an eval set, a model with an explanation, a query that returns in under a second.

Outside of building, I read papers on local LLM inference and write small validation tools.

FocusRAG, agents, evaluation, data engineering
LanguagesPython, SQL, Go, C/C++, R
Based inTallahassee, FL

"Not every complex problem needs more data. Most need a better system."


Resume

One page, AI Engineer track. Download it, or read the code on GitHub.


Get In Touch

AI Engineer and Forward Deployed Engineer roles, or a conversation about RAG and data systems.