Krishna Maddireddy

Krishna Maddireddy

Data Engineering, Analytics and AI, Los Angeles, CA

I've spent twenty years building data platforms from nothing. Now I build AI analytics products where every answer can be checked.

LinkedInGitHubEmail

Eight platforms built from zeroas first engineeras the data leaderas founder
  1. 2005Savings.comFirst engineer, original platform
  2. 2007LeadPointFirst data warehouse
  3. 2011BeachMintWarehouse and BI stack
  4. 2013FreedomPopData team, four countries
  5. 2019Just Slide MediaThe whole data function
  6. 2025PaakDataAI stock analytics
  7. 2026SpendQueryGovernment spending search
  8. 2026AskEiderSemantic-layer analytics

About

Data and analytics leader for e-commerce and consumer companies, hands-on in SQL, Python, BigQuery, Redshift and AWS. I own the whole stack, from KPIs and executive dashboards down to the pipelines underneath, and lately I've been building natural-language analytics, LLM-written analysis and MCP tools that let AI assistants query data directly.

Apps I've built

SpendQuery

Live, since 2026

Ask U.S. government spending anything, in plain English. SpendQuery turns the tens of millions of public records on USAspending.gov into answers: a sourced answer, a chart, and the SQL behind every number.

  • Natural language to a query plan, not to raw SQL: an LLM writes a structured QueryPlan and deterministic code turns it into SQL. The model never touches the database.
  • SpendQuery Chat and AI agents for contractors, each behind an evaluation gate; a golden set of 500+ questions reached 99.4% correct
  • Remote MCP server with 20 read-only tools and 4 prompts, with interactive tables and charts in Claude and ChatGPT
  • DuckDB plus dbt medallion pipeline holding 40M+ federal transactions (FY2023–2026), refreshed daily and reconciled against USAspending

PythonFastAPIDuckDBdbtNext.jsMCPClaudeAWSCloudflare

Read the full write-upVisit spendquery.comGitHub

PaakData

Live, since 2025

AI-first stock analytics for 12,000+ US stocks and ETFs. Ask in plain English and get exact screens back, with the filters shown so every answer can be checked.

  • Natural-language screening over hundreds of data fields
  • MCP server so assistants like Claude can screen stocks and pull company briefs directly
  • LLM and RAG pipelines that write company briefs with data-quality warnings built in, plus catalyst-based “similar to” search

PostgreSQLpgvectorClaudeOpenAIGeminiMCP

Visit paakdata.com

AskEider

Private preview, since 2026

Privacy-first natural-language analytics on a governed semantic layer. The model never sees your schema, values or rows; every answer comes from governed metric definitions.

  • A local resolver maps schema-free intent to metrics; MetricFlow compiles the SQL and DuckDB runs it read-only in a sandbox
  • Profiles any DuckDB, Parquet, CSV, SQLite, Postgres or MySQL source and generates a working dbt + MetricFlow layer
  • REST API, Python and TypeScript SDKs, MCP server, Slack and Teams apps; 450+ tests and an OWASP LLM Top 10 threat model

dbtMetricFlowDuckDBPythonTypeScriptMCP

GitHub

MLCompass

Building, since 2026

Bring a dataset; MLCompass recommends which kind of machine learning fits it, then trains, checks and explains the model end to end. Built for people who know their data but not ML, every result comes with a plain-language answer and a trust verdict.

  • Guide me turns four plain questions into a validated model. It ranks 10 ML approaches, applies leakage and baseline checks, explains predictions with SHAP and “what would change the outcome?”, and runs an Autopilot that keeps only gains that hold up under cross-validation.
  • Marketing measurement: a marketing mix model that captures paid media’s halo on direct and organic traffic (within 9.7% of known effects, against 12.3% for last-click), synthetic-control geo experiments, and Shapley + ML attribution that scales to 500K journeys.
  • Works with any AI model, or none: an agent that runs only tested tools and never generated code, and summaries written for executives, marketing, finance and sales. Exports a prediction API, a training script, and Snowflake and Databricks deployment kits, with 343 tests.

Pythonscikit-learnStreamlitFastAPIDockerLLMs

MCP Browser

Open source, MIT licensed, since 2026

The search engine that shows its work. An open-source AI search browser that sends each question to the few MCP servers that know the answer, chosen from the 48,000+ in the official MCP Registry, and shows several independent sources side by side instead of one generated answer. Runs locally with no account, and works without an AI key, with Ollama, or inside Claude.

  • Cross-check: checks a claim, or every fact in an AI answer, against independent sources and quotes each source's own sentence
  • Dependency check: reads a lockfile or a GitHub repo and compares two vulnerability databases
  • Tell me when: watches prices, releases and web pages, on your own computer
  • 106 tested pipelines that chain MCP servers together

Node.jsElectronMCPOllama

GitHub

Experience

  1. Dec 2019 to May 2026

    Head of Data Engineering and Analytics

    Just Slide Media, Los Angeles, CA

    Built the data and analytics team from scratch for a digital marketing company serving e-commerce clients. Owned strategy, hiring, vendors and delivery.

    • Combined Google Ads, Meta, TikTok, Amazon, Shopify, CJ, Impact Radius and other networks into one data lake
    • Connected online ad and e-commerce data with offline sales to measure the full funnel
    • Built multi-touch attribution models that cut wasted ad spend and lowered acquisition cost
    • Executive dashboards in Amazon QuickSight for marketing performance, conversion and revenue

  2. Jan 2025 to Present

    Founder, product and developer (part-time)

    PaakData

    Founded and built alone an AI-first stock analytics SaaS covering 12,000+ US stocks and ETFs.

    • Natural-language screening engine that shows the filters it applied
    • Automated prompt pipeline that tests new prompts against live data and publishes only the ones that pass
    • Prototyped AI analyst agents for 28 investing styles, each with its own screens and risk checks

  3. Jan 2013 to Nov 2019

    VP of Data Engineering, direct-to-consumer digital brands

    FreedomPop, Los Angeles, CA

    Built and led data engineering and analytics for a consumer business selling online in the US, UK, Mexico and Spain, across FreedomPop, Unreal Mobile and later Boost Mobile / Boost Infinite.

    • Full-funnel KPIs: acquisition, activation, billing, retention, churn, lifetime value and revenue per user
    • Combined 20+ sources, including SAP, CRM, payments, marketing networks and Firebase, into one data lake
    • Supply chain and revenue-checking reports for Finance and Operations
    • Set data quality, governance and reporting standards on Redshift, BigQuery, Pentaho and Airflow

  4. May 2012 to Jan 2013

    Senior Data Architect

    Machinima, Los Angeles, CA

    Led data and analytics work on marketing, customer acquisition and financial results for a large digital media network; designed data marts, ETL and executive reporting.

  5. Feb 2011 to May 2012

    Tech Lead, Business Intelligence

    BeachMint, Los Angeles, CA

    Built the data warehouse and BI stack (MySQL, Vertica, Pentaho) for a fast-growing online retailer with several subscription brands.

    • OLAP cubes and executive dashboards for sales, cohorts, customer behavior and fraud
    • Processed terabytes of clickstream and sales data for conversion and merchandising analysis

  6. Nov 2007 to Oct 2010

    Manager of Engineering, Data Warehouse

    LeadPoint, Los Angeles, CA

    Built the company's first data warehouse for an online lead marketplace, with data marts for pricing, quality and performance. Hired, mentored and led the team.

  7. Jan 2005 to Oct 2007

    Senior Software Engineer

    Savings.com, Los Angeles, CA

    First engineer. Designed the original architecture for the coupons and deals site and helped it grow to a $100M acquisition by Cox Target Media.

  8. 1996 to 2005

    Software Engineer

    Cisco, TIBCO Finance and Wipro Systems

    Enterprise software for finance and networking.

Education

  • M.S., Computer Applications
    National Institute of Technology, Karnataka
  • B.S., Mathematics
    Nagarjuna University

Articles

  1. Most federal notices about cybersecurity don't say "cybersecurity"

    How we rebuilt SpendQuery's topic search with AI at night and a single AI call per search, for about $3.50 of model time.

    , 5 min read

  2. We let an AI decide what a government grant is about

    How SpendQuery uses a narrow AI judge to tag federal awards by topic, while every dollar figure still comes straight from the records.

    , 1 min read

  3. Building Natural Language Stock Search Without Model-Generated SQL

    How PaakData turns plain-English stock questions into checked database queries, with a model that never writes SQL.

    , 6 min read

  4. Let AI Answer Data Questions — Without Ever Showing It Your Database

    A two-path architecture for natural-language analytics, in a world where the schema is the secret and wrong numbers don't crash, they get put in slides.

    , 5 min read

All articles

GitHub

  • KrishnaMaddireddy/fde-field-program

    Python

  • KrishnaMaddireddy/mcp-browser

    The search engine that shows its work: an AI search browser that routes questions to MCP servers, cross-checks claims across independent sources, audits dependencies and watches things for you. Runs locally, no account.

    JavaScript

  • KrishnaMaddireddy/agent-debate

    Several models answer one question independently, critique each other and rebut; code audits the evidence and measures what moved. Zero dependencies.

    JavaScript

Contact

The fastest way to reach me is krishmrm@gmail.com, or on LinkedIn.