Aakash Patel
LinkedIn

Senior Data Engineer · Calgary

I build data systems that have to be right.

I design and operate data platforms and production AI systems where correctness, latency, reliability and cost are part of the engineering problem.

Cuesta Partners · previously Citi and Avenue Living

View selected work LinkedIn
  1. Sources
    Operational systems, documents, events
  2. Ingestion
    Batch + streaming pipelines
  3. Transformation
    Modeled, tested, versioned
  4. Validation
    Reconciliation + quality checks
  5. Serving
    Warehouses, APIs, downstream users
  6. AI / Analytics
    Grounded in data that is right
How I think about a data system

01 · Data Platform

35% faster pipelines. ~20% lower compute cost.

Redesigned orchestration and compute across Microsoft Fabric and Databricks for a production data platform serving finance and operational workloads.

35%

faster execution

~20%

lower compute cost

50+

stakeholders

View case study
  1. Sources
    ERP / leasing / finance systems
  2. Fabric ingestion
    Scheduled + event-driven loads
  3. Delta / OneLake
    Single governed storage layer
  4. Transformation / Spark
    Simplified orchestration
  5. Quality + reconciliation
    Checks before anything ships
  6. Analytics / downstream users
    50+ stakeholders
Data platform — simplified architecture

02 · Production AI

AI answers are easy. Trust is the engineering problem.

An enterprise assistant answering tax and HR questions from internal documents. Retrieval quality, re-ranking, grounding and evaluation mattered more than the chat interface.

View case study
  1. Documents
    Tax + HR policy corpus
  2. Chunking / indexing
    Semantic embeddings
    retrieval quality
  3. Azure AI Search
    Hybrid retrieval
  4. Re-ranking
    Precision before generation
    grounding
  5. LLM orchestration
    LangGraph pipeline
  6. Evaluation
    Hallucination checks pre-ship
    hallucination checks
  7. Answer
    Cited, grounded, measured
RAG pipeline — retrieval before generation

03 · Real-time AI

In voice systems, latency is product quality.

Speech recognition, inference and synthesis were orchestrated asynchronously because every unnecessary pause makes the product feel broken.

View case study
  1. User speaks
    0 ms
  2. speech recognition
  3. partial transcript
  4. LLM inference
  5. first tokens
  6. speech synthesis
  7. audio response
Voice pipeline — stages overlap instead of waiting

Engineering Notes

01A $0.03 discrepancy took down a reconciliation pipeline.Reconciliation · Sep 202602When an AI agent says the data is "clean enough."Data quality · Sep 202603How we cut a scraping bill by 73%.Infrastructure · Oct 2026

View all notes

Experience

  1. 2026 — now

    Cuesta Partners

    Senior Data Engineer

    Data platforms and engineering systems across consulting engagements.

  2. 2025 — 2026

    Avenue Living

    Data Engineer

    Fabric, Databricks and production AI for finance and operational data.

  3. 2022 — 2025

    Citi

    Data Engineer

    Distributed Azure data pipelines for financial datasets.

Education — MSc Big Data Analytics, Trent University · BE Computer Engineering, GTU

About

I care about the parts of engineering that only become visible once people depend on a system: reconciliation, failure recovery, schema changes, observability, latency and cost. I work across data platforms and production AI, with an emphasis on systems that remain understandable and reliable after they ship.

Based in Calgary. Currently at Cuesta Partners; previously Citi and Avenue Living.

Have a hard data problem?

aakash98.pt@gmail.com

LinkedIn GitHub Calgary, Canada