Hi, I’m Shaswot.
From model experiments to reliable systems. I’m a computer engineering student building at the intersection of AI, data, and software.
A little more about me — download résumé
I work across agentic AI, machine learning, and backend engineering. I care about the details that turn an experiment into a useful system: good context, clear workflows, and thoughtful evaluation.
A few projects that show how I approach models, workflows, and the software around them.
An HR platform that brings policy questions, leave workflows, recruiting, and candidate evaluation into one system. Specialist agents work behind a shared supervisor, with hybrid retrieval for grounded answers.
Personalized mock interviews with timed questions, guarded session states, and evidence-based feedback. An eight-state workflow keeps the session predictable while two-stage evaluation separates scoring from coaching.
A credit-default experiment on 30,000 client records, connecting feature engineering and ensemble models to an API and risk dashboard. Evaluation focuses on the trade-offs of an imbalanced dataset.
Rainfall forecasting in Kathmandu using multi-year weather data, temporal and geographic features, and comparative machine-learning and deep-learning models. Related research was listed at ICETID 2026.
Speech, software, and systems in progress.
English/Nepali ASR experiments with PyTorch and NeMo. Public contributions include implementation documentation and a proposed set of fine-tuning notebooks.
View contributionA QR-based menu and ordering system with a Next.js frontend, a typed Fastify backend, and shared packages. An exploration of software beyond AI.
View repositoryAn early exploration of a local-first network intelligence agent. The planned pilot covers traffic visibility, anomaly investigation, and understandable recommendations.
View repositoryA few recurring ideas across my project work.
Retrieval, ranking, and useful evidence are the foundation of grounded answers.
HYBRID SEARCH / RERANKINGState machines, typed schemas, and guarded transitions make probabilistic systems easier to reason about.
STATE / VALIDATION / CONTROLModel evaluation, scorecards, and experiment tracking make improvements visible and comparisons meaningful.
EVALUATION / EXPERIMENTSRetries, background processing, and observability are part of building useful AI services.
RECOVERY / TRACING / TESTSShaswot Poudyal, Saurav Katwal, Rajad Shakya
PUBLIC CONTRIBUTIONS
FastConformer and IndicConformer implementation details, ONNX quantization, and robust figure handling.
Dual_Conformer_LateXAn implementation subsection and a figure for word error rate in the collaborative report.
Dual_Conformer_LateXProposed fine-tuning notebooks for a collaborative Nepali audio pipeline; not yet merged.
Nepali-audio-to-audio-pipelineContribution status checked September 2026.
The tools I use to move between experiments and applications.
LangChain · LangGraph · MCP · RAG · Vector databases
PyTorch · scikit-learn · XGBoost · Optuna · ASR · ONNX
Python · SQL · FastAPI · Pydantic · PostgreSQL · pgvector · Redis · Pandas · NumPy · MinIO · Alembic
MLflow · OpenTelemetry · Arize Phoenix · Docker · GitHub Actions
Tribhuvan University · Thapathali Campus
Exploring the building blocks of a minimal AI agent.
Multi-model streaming chat with local SQLite observability.
Document-grounded PDF question answering with RAG.
Collaborative and content-based book recommendations.
A focused retrieval experiment about Carlos Alcaraz.
ALSO IN THE CV Document Reconstructor System — OCR and object detection for reconstructing charts and tables from scanned documents. Read the résumé
Interested in AI systems, ML engineering, or collaborative technical work? Let’s connect.