Vibhu Mishra

Vibhu Mishra

  • Data engineer
  • Backend developer
  • Active open source contributor

I build data pipelines and the services that sit on top of them. At 314e Corporation I load and clean SQL Server data with SSIS and T-SQL, and I'm moving that ETL to Microsoft Fabric. Before that I built PySpark pipelines on Azure Databricks at Sigmoid. I also build RAG pipelines and MCP servers that let LLMs work with that data. Based in Bengaluru, open to remote work.

Email me GitHub LinkedIn

Career so far, drawn as a pipeline run: career_pipeline

2 success1 running
  1. btech_iiit_surat

    B.Tech, IIIT Surat

    Nov 2022 – Jul 2026

    • LeetCode 1809, CodeChef 3★, Codeforces Pupil
    • 500+ problems solved
    • Represented IIIT Surat in football at the Inter-IIIT Sports Meet twice (2024–2025)
    success
  2. intern_sigmoid

    Data Engineer Intern, Sigmoid

    Jan 2026 – Jul 2026

    • PySpark ETL on Azure Databricks, Bronze, Silver and Gold layers on ADLS Gen2
    • Scheduling and deploys with ADF, Airflow and GitHub Actions
    • FastAPI and Flask data APIs for analytics teams
    success
  3. associate_swe_data_314e

    Associate Software Engineer (Data), 314e Corporation

    Aug 2026 – now

    • Incremental loads from SQL Server and flat files with SSIS
    • T-SQL procedures and views, tuned with execution plans and indexes
    • Moving SSIS ETL to Fabric pipelines and Lakehouse tables
    running

Data pipelines

Batch pipelines that pull data from public sources, model it, and check it before anyone queries it.

Crypto Market Data Pipeline

Medallion lakehouse on Databricks

Team project, 2026 Source on GitHub

Loads live and historical market data for the top 20 coins from the CoinGecko API into Delta Lake, and models it into tables a trading dashboard reads.

LineageDelta tables by layer
  1. CoinGecko APIsource
    Markets, plus 30 days of 4-hour OHLC candles
  2. Bronzeraw, merged
    bronze_market_databronze_ohlc_data
  3. Silverstar schema
    dim_coindim_datesilver_market_metricssilver_ohlc_metrics Rows that fail checks go to silver_market_quarantine
  4. Goldtrader views
    gold_market_snapshotgold_ohlc_historygold_performance_metricsgold_market_summary
  • Ingestion retries with exponential backoff and stays inside CoinGecko's rate limit.
  • Loads use Delta MERGE, so re-running a batch doesn't duplicate rows.
  • Silver separates dimensions from facts. Rows that fail quality checks are quarantined instead of dropped, and every run is logged with its status and row counts.
  • Gold adds support and resistance levels, performance metrics and a recommendation score for a Databricks Lakeview dashboard.

Stack: PySpark, Databricks, Delta Lake, Python, CoinGecko API

Interop Mapper

ETL and search for the FHIR health data standard

Solo project, 2026 Source on GitHub

FHIR is the HL7 standard that health systems use to exchange records. Type a field like “date of birth” and Interop Mapper shows where it goes (Patient.birthDate) and how it changed between FHIR R4, R4B and R5, with no internet needed once loaded.

Lineagebuild steps, in order
  1. HL7 registryfetch_packages.py
    R4, R4B and R5 core, plus the Extensions Pack
  2. SQLite + FTS5build_index.py
    24,754 elements2,196 extensions4,081 search params
  3. Static exportexport_web.py
    JSON per version, and R4 to R4B to R5 diffs computed element by element
  4. Offline appweb/
    10,486 examples validatedone 5 MB HTML file
  • A Python ETL pulls the official HL7 packages and loads every element, extension and search parameter into SQLite with full-text search.
  • Version diffs cover added and removed fields, cardinality, type and binding changes, and flag likely renames.
  • Generates example JSON, Python, JavaScript and curl for every resource and field. All examples validate against the independent fhir.resources models.
  • Ships as static files: an installable offline app, or a single HTML file you can email.

Stack: Python, SQLite FTS5, JavaScript, service workers

Backend services

APIs and server code, and the data models behind them.

Vibnex

URL shortener API

Solo project, 2025 Live site Source on GitHub

Paste a long link, pick a short code, and share the short one. Two route handlers do all the work: one creates links, one redirects.

Request logexample requests
  1. POST/api/create200 OK{"long": "github.com/vibhuthespeedst", "code": "gh"}
  2. POST/api/create409 Conflict{"long": "example.com", "code": "gh"}
  3. POST/api/create400 Bad Request{"long": "not a url", "code": "x"}
  4. GET/gh302 FoundLocation: https://github.com/vibhuthespeedst
  5. GET/nope404 Not Found
  • The create route validates input, adds https:// when the scheme is missing, and checks the short code is free before inserting.
  • The redirect route does a single lookup by short code and answers with a 302.
  • Moved storage from SQLite to Supabase PostgreSQL so it runs on Vercel's serverless functions.

Stack: Next.js route handlers, Supabase (PostgreSQL), Vercel

FinTrack

Personal finance backend and app

Team project, 2025 Source on GitHub

Tracks spending across several accounts. Upload a bank statement PDF or a receipt, and its transactions are read, categorized and saved.

Data modelPrisma, PostgreSQL

User

  • id pk
  • clerkUserId unique
  • email unique

Account

  • userId fk User
  • type enum
  • balance decimal
  • isDefault bool

Transaction

  • accountId fk Account
  • userId fk User
  • amount decimal
  • category, date
  • recurringInterval enum

Budget

  • userId fk User, unique
  • amount decimal
  • lastAlertSent time
  • Server actions validate input with Zod, then write the transaction and update the account balance in one Prisma $transaction, so balances can't drift from their rows.
  • Gemini reads uploaded statements and receipts and returns structured transactions, categorized from their descriptions.
  • Bulk delete reverses each transaction's effect on its account inside the same database transaction.
  • Filtered views export to .xlsx. Clerk handles sign-in.

Stack: Next.js server actions, TypeScript, Prisma, PostgreSQL (Supabase), Zod, Gemini API, Clerk

LLM apps

LLMs on top of real data, with retrieval and tool calling so the answers stay grounded.

DataWiz

Chat with a CSV or Excel file

Solo project, 2025 Live app Source on GitHub

Upload a table and ask questions about it in plain English. Answers are computed from the data, not guessed, and each one shows how it was worked out.

One questionfrom question to answer
  1. Questionuser
    “Which region has the highest total revenue?”
  2. RetrieveTF-IDF
    revenue profileregion profile8 matching rows
  3. Tool callpandas
    aggregate: sum(revenue) by region
  4. AnswerGemini
    Written from the tool result, with every step shown under it
  • RAG: every column profile and row is indexed with TF-IDF, and each question pulls the most relevant ones into the prompt.
  • Gemini calls four pandas tools (aggregate, filter rows, value counts, correlation). The math runs in pandas, and nothing the model writes is run as code.
  • Also profiles the table, draws eight kinds of charts, and filters and exports rows.
  • Tested with a scripted stand-in for Gemini, so the tool-calling loop is checked without an API key.

Stack: Python, LangChain, Gemini API, scikit-learn, pandas, Streamlit, Plotly

Skills

Data engineering

Processing
PySpark, Spark SQL, T-SQL, Pandas
Platforms
Azure Databricks, Delta Lake, Microsoft Fabric, Azure Synapse, Snowflake
Orchestration
Azure Data Factory, Apache Airflow, SSIS, Kafka
Modeling
Medallion layers, star schema, SCD Type 2
Data quality
PII masking, validation and reconciliation, lineage

Backend

Languages
Python, SQL, Go, Rust, TypeScript
Services
FastAPI, Flask, Next.js route handlers, REST APIs
Databases
SQL Server, PostgreSQL, SQLite, Prisma
Infrastructure
Docker, Kubernetes, Terraform, GitHub Actions, Azure, GCP
LLM apps
RAG pipelines, MCP servers, LangChain, LangGraph

Contact

I'm looking for remote data engineering and backend work. Email is the quickest way to reach me.