Miraj Pudasaini
CV
Open to AI engineer roles

Miraj Pudasaini

AI engineer in London. I build RAG agents and hybrid search over financial time series, and the data pipelines that feed them.

At I Tech Care I run the Airflow and Databricks pipelines that load market data into Azure, and I have shipped vertical AI agents that answer questions over EODHD tickers and a BM25 and embedding search across millions of financial series. Before that I spent two years at Zola Analytics building its macro data platform and an agentic RAG that creates charts from it.

View workDownload CV
Or email [email protected]

Projects

Each project states the problem, what I built and a measured result.

ContextTrace

Semantic search over GitHub discussions

A retrieval service that answers questions about a repository with passages linked to the source issue or pull request comment. An incremental, idempotent ETL on Cloudflare Queues with deterministic passage ids and content hashes feeds Workers AI embeddings into Supabase pgvector. The whole service runs on free tiers with budget ledgers, job leases and retries, and a 4 to 6x cut in transform CPU.

Hit@3 0.65 and thread Hit@5 0.95 on a frozen 25-question set; keyword baseline 0.05

  • TypeScript
  • React
  • Cloudflare Workers
  • Workers AI
  • Supabase
  • pgvector
Code Live demo

OpenAQ Data Platform

ETL platform for live air-quality data

Airflow DAGs pull measurements from the OpenAQ API into PostgreSQL through a repository and unit-of-work layer, with pytest coverage and a Docker Compose stack for the whole platform.

  • Python
  • Airflow
  • PostgreSQL
  • Docker
  • pytest
Code

Frameprov

Sentiment over live tweets

Pulls tweets on a chosen topic through the Twitter API and scores how many of a 100-tweet sample speak positively about it, using NLTK.

  • Python
  • NLTK
  • Twitter API
Code

Skills

What I use at work and in my own projects.

AI and ML

  • RAG agents
  • Hybrid search
  • BM25
  • Embeddings
  • Titan
  • Workers AI
  • pgvector
  • Retrieval evals
  • nDCG@10
  • Hit@k
  • MRR
  • LLM APIs

Data

  • Python
  • SQL
  • Airflow
  • Dagster
  • Databricks
  • Spark
  • Azure
  • AWS
  • Snowflake
  • PostgreSQL
  • TimescaleDB
  • Neo4j

Software

  • TypeScript
  • React
  • Next.js
  • Node.js
  • Java
  • Spring Boot
  • Docker
  • Cloudflare Workers
  • Supabase
  • Serverless CI/CD

Experience

What I did in each role, with the outcome.

  1. Forward Deployed Full-stack Engineer

    I Tech Care

    June 2025 – present

    • Design and run Airflow pipelines with Databricks and Spark connectors that load prices and market tickers from major stock exchanges into Azure Data Lake Storage.
    • Built vertical AI agents that run RAG over EODHD time-series tickers to automate financial market analysis.
    • Led a hybrid BM25 and Titan-embedding search over millions of financial series, gated on nDCG@10 and P@12.
    • Deployed data infrastructure for quality analysis and ingestion that raised the data team’s productivity by 40%.
    • Airflow
    • Python
    • SQL
    • Databricks
    • Spark
    • Azure

  2. Data Engineer

    Zola Analytics

    Feb 2023 – May 2025

    • Replaced Excel workflows with Dagster and Python pipelines, raising ingestion frequency by 60% and improving monitoring and alerting across the data process by 80%.
    • Built API clients that migrated the OECD, IMF, BIS and FRED databases into a dimensional time-series model on the Zola data platform.
    • Optimised the Python, SQLAlchemy, Neo4j and TimescaleDB platform, cutting transaction time by 70% and the monthly AWS storage bill.
    • Worked with the ML and AI team to design and deploy an agentic LLM that runs RAG over macro time series to create charts automatically.
    • Dagster
    • Python
    • Pandas
    • SQL
    • TimescaleDB
    • Neo4j
    • Snowflake
    • AWS

  3. Full-stack Software Engineer

    Green Codes

    April 2019 – June 2020

    • Modernised a legacy Java CRM to Node.js, React and Next.js, making the CRM and dev tools 12% faster.
    • Moved the AWS CI/CD pipeline to serverless, cutting its operating cost by 20%.
    • Tuned SQL queries so user retrieval used 20% less database resources.
    • Java
    • Spring Boot
    • Node.js
    • React
    • Next.js
    • AWS

Education

  1. 2021 – 2023

    MSc Computer Science (Upper Second Class)

    York St John University, London

  2. 2017 – 2020

    BSc (Hons) Computer Science (First Class)

    University of Wolverhampton at Herald College, Kathmandu

Hiring an AI engineer who ships retrieval systems with evals, and the pipelines behind them?

Email meDownload CV