ASHISH TREHAN

ashishtrehan10@gmail.com  •  706-294-0985  •  www.ashishtrehan.com

Data Scientist & AI Engineer  •  Agentic AI & Orchestration  •  Data Engineering & Analytics

SKILLS & SUBJECT MATTER EXPERTISE

Agentic AI & Orchestration: LangChain, LangGraph, LlamaIndex, n8n, Airflow, AutoGen, Multi-Agent Systems, Tool/Function Calling, RAG Pipelines, Agent Memory & State Management

AI & Computer Vision: LLMs (Claude, GPT-4, Llama), YOLOv8, Prompt Engineering, Fine-tuning, Hugging Face, NLP (spaCy, NLTK)

Machine Learning: Classification, Prediction, Clustering, Optimization (sklearn), Recommendation Systems (Collaborative Filtering & Knowledge-Based), Meta-Heuristics (Genetic Algorithm, PSO, Simulated Annealing), Statistical Modeling (Statsmodels, R, SAS)

Data Engineering: Python (NumPy, SciPy, Pandas), SQL (Postgres, MySQL, SQL Server, Snowflake, MariaDB), dbt, Airflow, Flask, FastAPI, Snowpipe, Snowpark, AWS (Lambda, ECR, Kubernetes), Airflow, Databricks, Azure Full Stack, Google Cloud, Redis, Alteryx, RabbitMQ, Alooma, Docker, Git, Codefresh, CircleCI

Data Visualization: Plotly, D3.js, Tableau, PowerBI, Periscope, Looker, Streamlit

Optimization: Pyomo, PuLP, Google OR-Tools

Industries: Automotive, Logistics, Healthcare, Security, Geospatial, Media, Advertising, EdTech, Procurement, Pharma, IoT, Insurance

WORK EXPERIENCE

SELF-EMPLOYED — CLIENT WORK

SmartSense May 2025 – Present

AI & Data Engineering Consultant Atlanta, GA

dbt / Snowflake Data Pipeline

• Designed and deployed a multi-layer dbt pipeline (bronze → silver → gold) in Snowflake, transforming raw IoT sensor readings into a production-ready asset failure scoring table through 20 modular, dependency-managed models.
• Engineered a complex feature-label construction model computing days-to-next-failure, days-after-last-failure, and multi-horizon failure indicators (7d/14d/21d) against a live failure incident dimension, resolving SCD fan-out via window functions to maintain ML grain integrity.
• Implemented an incremental dbt model with merge strategy and SHA-256 surrogate key generation for the ASSET_FAILURE_SCORE table, enabling efficient daily upserts while preserving historical scores and model metadata (precision, recall, F1).
• Architected a multi-environment dbt profile structure (QA / staging / prod) backed by Snowflake env-var injection, enabling the same containerized image to target any environment without code changes.
• Built dual-model scoring output surfacing both logistic regression and LightGBM failure probabilities (0–100) alongside heuristic severity-tier warnings (SEV1/SEV2/SEV3) into a single unified table consumed by downstream analytics.

ML Training Pipeline (Python / Snowflake / sklearn)

• Built an end-to-end asset failure prediction training pipeline in Python, orchestrating Snowflake ingestion, temporal train/eval/test splitting, feature scaling, and L1-regularized logistic regression model fitting with no manual data movement.
• Engineered a config-driven data ingestion layer against Snowflake using dynamic query construction — parameterized by lookback windows, column selection, and filter predicates — with support for chunked fetch_pandas_batches to handle large datasets efficiently.
• Designed a temporal train/eval/test split strategy with configurable window sizes and exclusion buffers, enforcing strict chronological boundaries to prevent data leakage in time-series failure prediction.
• Implemented a pluggable time-based sample weighting system supporting exponential, inverse, linear, and quadratic decay functions keyed on days-to-next-failure, allowing the model to upweight imminent failure signals during training.
• Migrated pipeline configuration from ad-hoc Python constants to a Pydantic-validated YAML schema, enforcing type safety across all training hyperparameters, Snowflake source definitions, column mappings, and evaluation thresholds.

Infrastructure & Developer Tooling

• Containerized the full training and dbt pipeline stack using Docker and Docker Compose, with environment-specific Snowflake credentials injected via .env file mounting and a single shared profiles.yml eliminating auth drift across services.
• Resolved a multi-package pytest collection conflict in a monorepo spanning a FastAPI service, dbt project, and ML training package — introducing a root pytest.ini with scoped testpaths and norecursedirs to prevent third-party dbt package tests from shadowing project-level conftest resolution.

Nimbus November 2024 – Present

AI & Data Engineering Consultant Atlanta, GA

• Designed and deployed an agentic video analysis pipeline using FastAPI and YOLOv8, orchestrating multi-step ML inference, transcoding, and delivery workflows end-to-end with no manual intervention.
• Engineered an LLM-powered agent with tool-calling capabilities to query processed video metadata, enabling natural language Q&A over object detection results stored in AWS S3.
• Architected an end-to-end agentic data workflow integrating AWS S3 for secure ingestion and delivery, achieving seamless transition from raw footage to processed and queryable results.
• Containerized the entire agentic stack using Docker and Docker Compose, ensuring reproducible, scalable deployments across cloud infrastructure.
• Implemented automated video transcoding via FFmpeg as an orchestrated step in the pipeline, converting ML inference outputs into web-compatible MP4 formats.

Big Health 2025

AI & Data Engineering Consultant Remote

• Built and maintained FHIR-based healthcare integrations with Redox to ingest clinical service requests securely and reliably.
• Orchestrated multi-system data flows using n8n workflows, automating invoice submission to insurers and connecting disparate healthcare data sources — a key agentic orchestration pattern applied to clinical operations.
• Developed Django REST endpoints with comprehensive automated test coverage to ensure stability and compliance.
• Built Looker dashboards and dbt pipelines to support the Analytics team; constructed datacubes consumed by Product and Marketing.

SpendHQ September 2023 – October 2024

Tech Data Lead Atlanta, GA

• Mentored five junior Data Engineers through weekly 1:1s, technical reviews, and career development coaching.
• Agentic & LLM Development: Built an internal mock data generator leveraging LLMs with structured tool calling and time series modeling to produce synthetic, industry-specific datasets — enabling Sales Consulting to demo the product to OEM, CPG, and other verticals without exposing real client data.
• Co-architected and deployed a new normalized data model in Snowflake (query/transform layer) with SingleStore as the read layer, replacing a legacy black-box BFT architecture.
• Migrated ETL and Rules execution from Alteryx/Assembler to a self-service solution using Snowflake Stored Procedures/Tasks and FastAPI; deployed Redis to push events on Rules Manager status changes.
• Introduced Flyway for migration deployment and incorporated Lexorank within the data pipeline for dynamic dataset reordering.

HealthgradesJuly 2021 – August 2023

Lead Senior Data Engineer Atlanta, GA

• Mentored two BI Analysts and two Data Engineers through weekly 1:1s and hands-on technical guidance.
• Agentic Data Workflows: Replaced manual email-attachment download workflows with event-driven Lambda + Airflow agents that autonomously hit APIs or pull data from external cloud providers, eliminating human-in-the-loop data ingestion steps.
• Architected and deployed 60 Snowpipe workflows to ingest from S3 to Snowflake, sunsetting Trifacta and MySQL and saving the team $100,000 in annual renewal costs.
• Co-architected the mapping of claims data to FHIR standard leveraging Apigee, Azure Data Factory, and Databricks, enabling cross-insurer data portability in compliance with regulatory requirements.
• Managed AdRevOps data migration from MySQL/SQL Server to Snowflake following Data Mesh principles for self-service, domain-oriented governance and Data-as-a-Service.
• Designed Data Dictionary and Catalog in Alation for PII/PHI data; architected RBAC within Snowflake for staging tables and governed views.
• Deployed a Streamlit POC to replace Tableau dashboards for Ad Rev Ops, enabling daily real-time reporting from weekly cycles; built a predictive/time-series impression and click forecaster with 5–8% sMAPE error.

Uplift K12 December 2020 – Now

Data Scientist | Boardmember Atlanta, GA | Houston, TX

• Designed and deployed new data architecture on AWS for both the legacy product and new product Nora; built and deployed the new product API using FastAPI.
• Implemented algorithmic ranking for incoming requests using optimization and heuristic approaches on the new product platform.
• Served as Technical Project Manager: scored and prioritized engineering tasks, managed sprint planning in Aha/Asana, and introduced CircleCI for continuous integration.
• Wrote all regression and acceptance tests for legacy product lines using Selenium; tested BI reporting across Tableau, Periscope, and Looker.

Clutch Technologies July 2016 – September 2020

Data Scientist | BI Manager Atlanta, GA

• Agentic Recommendation & Assignment System: Designed a multi-stage agentic pipeline that (1) generated vehicle recommendations via Matrix Factorization Collaborative Filtering, (2) solved a combinatorial optimization problem to match users to available vehicles, and (3) assigned concierges for pick-up/delivery — automating 81% of all incoming requests, up from 40%.
• Built all algorithmic services (Recommender, Assignment) as backend APIs in Flask/FastAPI with nosetests-based unit testing to ensure global optimization quality.
• Designed and maintained 30 client-facing dashboards (168 visualizations) in Periscope for Fortune 500 OEM and regional dealership clients.
• Built a data pipeline for telematics (geolocation) data: Elasticsearch → AWS Glue → Redshift → Periscope/Kibana.
• Used Segment and Redis to build marketing segmentation pipeline enriching web traffic IP data with zip codes; channeled all data into Snowflake.

dRISTi 360 July 2015 – July 2016

Senior Associate, Data Scientist Atlanta, GA

• Advanced predictive modeling within the AdTech stack by leveraging social stream data from Twitter and Facebook for audience segmentation strategies.
• Co-designed an Audience Insights application surfacing industry KPIs and data visualizations, reducing individualized client presentation build time by 50%.
• Built a text classification pipeline (Random Forest + SVM) improving tweet classification accuracy from 71% to 85%; used SVD + t-SNE for interactive keyword discovery.

Slalom Consulting June 2014 – June 2015

Junior Data Scientist Atlanta, GA

• Built SQL workflows in AWS Redshift and Tableau dashboards for a cross-customer cable client service platform, combining call center, web stream, and field data.
• Deployed Survival Analysis on worker retention identifying key attrition milestones; deployed an internal self-service Python/R server for the Analytics practice.

Sogeti Consulting (CapGemini) March 2013 – July 2014

Practice Area Lead — Data Analytics Atlanta, GA

• Led the first Big Data initiative for NCR IT Services using Teradata Aster with Random Forest ensemble models, improving ATM failure work order prediction from <10% to 32%.
• Used SAS, Chi-Squared tests, and ROC analysis to validate model performance vs. legacy rule-based systems.

EDUCATION

M.S. Applied Economics  —  University of Georgia, Athens, GA 2013

B.S. Architecture  —  University of Georgia, Athens, GA 2010