Open to remote roles · 6 years in data

Zakir Siddiqui

I work as a data scientistML engineerdata analystbackend developeragentic AI engineercivil engineer

I build Python and SQL pipelines, predictive models and Power BI dashboards. At Deccan Holdings they handled 4.5M records a day and cut a 6-hour report to 45 minutes.

Results in numbers

  • 4.5Mrecords a day through my pipelines
  • 6h→45mmorning report, before and after
  • 82.1%of churners caught by the churn model
  • 7.47×smaller PyTorch model, 0.9973 cosine
  • 450K+video frames quality-checked
  • ₹38Lrecovered from infrastructure billing

The story · 2016 → now

Civil engineer turned data engineer.

  1. 01 · 2016

    B.E. in civil engineering

    Jawaharlal Nehru Engineering College (MGM). Structural design, estimation and surveying: my first models were made of concrete.

  2. 02 · 2017 → 2019

    On site, with the drawings

    Intern at CBRE, checking sites against bar bending schedules and drawings. At Teemage Builders I made precast tracking about 20% faster.

  3. 03 · 2018

    NICMAR, construction management

    A PG diploma: scheduling in Primavera P6, bills of quantities, QA and QC. I still read a BOQ as well as a query plan.

  4. 04 · 2019 → now

    The building became data

    BI and MIS at Deccan Infrastructure recovered INR 38 lakh from billing. Then healthcare, finance and AI: pipelines, models and dashboards.

What I do · four disciplines

From raw rows to shipped systems.

01

Data analysis

SQL with CTEs and window functions, Power BI and DAX. Three dashboard suites for 8 fund managers.

  • SQL
  • Power BI
  • DAX
  • Excel
02

Data science and ML

Classification, regression, time series and NLP, tuned to the cost of a wrong call. PyTorch for quantization, video search and deepfake detection.

  • scikit-learn
  • XGBoost
  • PyTorch
  • spaCy
03

Backend and pipelines

Scheduled Python jobs that ingest, validate and publish, with tests and CI on every repository. Flask APIs, SQLite and Parquet stores.

  • Python
  • Flask
  • GitHub Actions
  • SQLite
04

Web development

HTML, CSS and React for pages like this one, and Streamlit apps on top of each model so people can actually use it.

  • HTML and CSS
  • React
  • Streamlit
  • Three.js

Lab · try it yourself

Play with the churn model.

customer-churn · test setinteractive
Churners caught307 / 374
Recall82.1%
F10.796
Net benefit$257,880

Each dot is one of the 374 customers who churned in the test set. Lit dots are the ones the model flagged in time. Lowering the cutoff to 0.38 trades a little precision for 37 more saved customers.

churn_by_contract.sql7,043 customers
SELECT contract,
       AVG(churned) AS churn_rate
FROM   customers
GROUP BY contract
ORDER BY churn_rate DESC;
Month-to-month42.7%
One year11.3%
Two year2.8%

3 rows · ran on 7,043 customers

Agentic AI · LLMs that use tools

Agents that plan, call tools and check their work.

I build LLM agents that break a goal into steps, call tools such as SQL, search and code, and check the result before they answer. Then I measure them: my eval harness gates every release, and at Outlier and micro1 I rate agent replies against rubrics.

  • Tool use and function calling

    Typed tools for SQL, search, files and APIs, with validation and retries.

  • Planning loops

    Plan, act, observe and repeat within a step budget, with memory between steps.

  • Retrieval (RAG)

    Grounding answers in documents and video: the video-search project finds the moment that matches a sentence.

  • Evals and release gates

    A 60-task agent benchmark with bootstrap intervals and McNemar tests. Candidate v2: pass rate 80.0% → 95.0%, p = 0.0039.

  • Guardrails and human review

    Rubric-based rating of LLM and agent replies in Hindi and English, with a written reason for each.

agent.run(goal)runs in your browser

    Real tools over this site's data. The planner here is a small rule-based router so you can watch every step; in production that role is played by an LLM.

    Selected work · 12 repositories

    Projects, with their test results.

    Every number below comes from the repository's own tests. Scroll sideways through them.

    All repositories on GitHub
    01Python · statistics · experimentation

    A/B testing platform

    CUPED cuts the sample an experiment needs by 40%, and a sequential SPRT lets people check results every day without inflating false positives. Tested on 300 simulated experiments.

    Sample saved
    40.0%
    Peeking error
    24.6 → 4.1%
    Simulations
    300
    See the experimentation code
    02Python · SciPy · Streamlit

    Marketing mix model

    Weibull adstock, Hill saturation and a non-negative Bayesian ridge, then SLSQP moves budget between channels: +1.09% weekly revenue for the same spend, over 10 simulated 156-week runs.

    R²
    0.9643
    MAPE
    1.64%
    Per year
    +$426,419
    See the marketing mix code
    03Python · BG/NBD · survival analysis

    SaaS lifetime value

    BG/NBD and Gamma-Gamma models forecast 12-month value for 1,500 accounts, with Markov multi-touch attribution and Kaplan-Meier retention curves.

    12-month CLV
    $13,418
    Top channel
    Sales demo
    Log-rank p
    < 1e-13
    See the CLV code
    04Python · operations research

    Supply chain inventory

    Multi-echelon safety stock for 1 warehouse, 4 hubs, 16 stores and 20 SKUs, plus a 52-week simulation of how order swings grow upstream.

    Saved a year
    $117,185
    Holding cost
    −7.04%
    Service level
    100%
    See the inventory code
    05Python · pricing · simulation

    Marketplace surge pricing

    Sets surge prices across 8 city zones and 96 fifteen-minute slots, and holds 1.0× when supply is balanced. Compared with flat pricing over 10 simulated days.

    Mean surge
    1.36×
    Bookings a day
    $2.82M
    Driver share
    75%
    See the pricing code
    06Python · scikit-learn · Streamlit

    Customer churn and retention

    A churn model on 7,043 telecom customers. Moving the cutoff from 0.50 to 0.38 catches 307 of 374 churners in the test set.

    Recall
    82.1%
    ROC-AUC
    0.848
    Retained
    $257,880
    See the churn code
    07Python · SQLite · Parquet

    Equity forecasting pipeline

    Forecasts for 120 tickers in a SQLite and Parquet store, with checks for splits, bad ticks and gaps. The morning report runs in under 45 seconds.

    Tickers
    120
    Windows
    10
    RMSE vs bench
    −0.6 to −0.8%
    See the forecasting code
    08SQL · DAX · Power BI

    Enterprise sales BI

    A star schema with 40,000 sales facts, 12 KPIs in SQL and DAX, and drill-down by time, region, product, segment and channel.

    Fact rows
    40,000+
    KPIs
    12
    Dimensions
    5
    See the BI code
    09Python · statistics · CI gate

    LLM agent eval harness

    A 60-task benchmark for LLM agents with bootstrap intervals, permutation and McNemar tests, and a CI gate that blocks a weaker release.

    McNemar p
    0.0039
    p95 latency
    680 → 578 ms
    Gate
    PASS
    See the eval harness
    10PyTorch · quantization

    Quant-ops engine

    INT8 per-channel and INT4 group-wise quantization for a 12.86M-parameter model, with an FP16 softmax overflow test.

    Compression
    7.47×
    Cosine sim
    0.9973
    Params
    12.86M
    See the quantization code
    11PyTorch · retrieval · re-ranking

    Video moment search

    Finds the moment in a video that matches a sentence: shot cuts, a 512-dimension dual encoder and a cross-attention re-ranker.

    p50, CPU
    26 ms
    p95, CPU
    31 ms
    Encoder
    512-d
    See the video search code
    12PyTorch · SE-ResNet · FFT

    Deepfake detector

    Spots AI-made faces by pairing an SE-ResNet image stream with the 2D FFT power spectrum, which generators get wrong.

    EER
    10.0%
    Precision
    1.000
    p50, CPU
    19 ms
    See the detector code

    Experience · since 2017

    Eight roles, four numbers they moved.

    • SLA compliance81%94.5%
    • Dataset rework4.8%1.2%
    • Morning report6 h45 min
    • Approval cycle5 days1.5 days

    beforeafter

    1. 2026

      AI training contributor, Outlier

      Hindi and English audio, TTS and agent replies rated against rubrics.

    2. 2026 → now

      AI data specialist, micro1

      Video annotation, rated senior.

    3. 2025 → 2026

      Data analyst and ML specialist, Deccan Holdings

      Pipelines for 4.5M records a day; a 6-hour report down to 45 minutes.

    4. 2023 → 2024

      Healthcare data analyst, eHealth Technologies

      12 US client networks, HIPAA audits.

    5. 2022 → 2023

      Operations data analyst, Webonise Labs

      450,000 frames checked.

    6. 2019 → 2022

      BI and MIS analyst, Deccan Infrastructure

      INR 38 lakh recovered from billing.

    7. 2018 → 2019

      Project operations, Teemage Builders

      Precast tracking, about 20% faster.

    8. 2017

      Civil engineering intern, CBRE

      Site checks against BBS and drawings.

    Toolbox · 14 tools, 4 languages

    The stack, from AutoCAD to PyTorch.

    Drag the nodes around. Pick a discipline to light up its tools.

    • PythonAdvanced
    • SQLAdvanced
    • Machine learningAdvanced
    • Power BI and ExcelAdvanced
    • AI rating and annotationAdvanced
    • AutoCAD, STAAD Pro, ETABS, RevitAdvanced
    • Primavera P6 and MS ProjectAdvanced
    • Git and GitHubIntermediate
    • HTML and CSSIntermediate
    • ReactBeginner
    • JavaBeginner
    • Hindi · English · Urdu · MarathiNative · C1 · fluent
    2016

    B.E. in civil engineering

    MGM's Jawaharlal Nehru Engineering College

    2018

    PG diploma, construction management

    NICMAR

    2025

    Data science and analytics

    Coding Ninjas certificate

    Terminal · SQL over my career

    Skip the resume. Query it.

    zakir@portfolio: ~/career.db

    Next · remote roles and contracts

    Have data that should be making decisions?