Skip to content
Open to roles — anywhere in the US

Atharva Pingale

>I buildagentic systems

I build production RAG and agent systems, and the evaluation that proves they work.

View work Get in touch

~/ United States · 4+ yrs · M.Eng CS · GPA 4.0

01about

Engineering AI that earns trust in production

I'm an applied AI/ML software engineer with 4+ years of professional experience. I build production RAG and agent systems, plus the evaluation that proves they work.

Since May 2024 I've owned the AI platform across client engagements, sometimes as the only engineer on a project. That covers multi-agent orchestration in LangGraph, memory over pgvector and a Graphiti knowledge graph, and LangSmith eval suites that gate merges in CI. The services ship as containerized FastAPI apps on AWS, built in CI and pushed to a container registry.

I've worked across healthcare (including HIPAA work), banking, manufacturing and warehouse operations, law-enforcement training, and consumer apps. Before this role I built a data platform and a Neo4j knowledge graph on Azure and Google Cloud.

I build reusable patterns and tooling so engineers without an AI background can extend what I ship, and I use AI coding tools daily to get from research to a working pipeline quickly.

now
Software Engineer, Big Kitty Labs
currently
Columbus, OH
open to
Anywhere in the US — relocation, hybrid, or remote
education
M.Eng CS · GPA 4.0 · Univ. of Cincinnati
domains
Healthcare (HIPAA) · Banking · Manufacturing & warehouse ops · Law-enforcement training · Consumer apps
tooling
Claude Code · Cursor · Copilot, daily
selected outcomes
Push
Not polling
Replaced fixed-interval REST polling with Kafka events over SSE, so results appear the moment they're ready
0%
Faster analytics
Redesigned a partitioning mismatch in the analytics layer
0.0%
Ship on A/A runs
upgrade-report, personal project: 500 simulated runs where nothing changed
1.3-1.6x
Faster prefill than Ollama
inferno, my Rust CPU inference engine, personal project: TinyLlama-1.1B, Q4_0, 8 threads
0.0
Graduate GPA
M.Eng Computer Science, University of Cincinnati
02experience

Experience & education

Four-plus years across agentic AI, distributed backends, and large-scale data platforms.

Work experience

  1. Software Engineer · Big Kitty Labs

    May 2024 — Present

    Columbus, OH

    I own the AI platform across client engagements, sometimes as the only engineer on a project: production RAG and agent systems, plus the evaluation that proves they work.

    • Architected multi-agent orchestration pipelines in LangGraph & LangChain with MCP tool integrations, structured output schemas, and safe tool execution for complex multi-step reasoning.
    • Built persistent memory over pgvector and Graphiti knowledge graphs, grounding LLM responses in domain data.
    • Stood up CI/CD for agents with automated LangSmith eval harnesses, gating task success, grounding, latency, and drift before merge.
    • Shipped enterprise safety guardrails and anti-jailbreak middleware with strict tenant isolation and audit-ready access logging.
    • Replaced REST polling with Kafka events streamed over SSE, so users see results as soon as they're ready.
    • Owned large React/TypeScript frontends (Redux, Zustand, Jotai) and the build/test tooling (Vite, esbuild, Vitest) across the codebase.
    LangGraphLangChainMCPpgvectorFastAPIAWSDockerKafkaReactTypeScript

  2. Software Engineer · Angular Minds

    Nov 2021 — May 2023

    Remote

    Built full-stack data platforms and AI-powered workflow tools for healthcare and enterprise clients on Azure and Google Cloud.

    • Built a full-stack data platform on Azure and Google Cloud.
    • Built a semantic knowledge graph in Neo4j linking entities across fragmented enterprise data.
    • Made critical analytics 40% faster by redesigning a partitioning mismatch in the analytics layer.
    • Shipped an automated data governance framework with audit trails and auto-correction.
    • Operated Airflow pipelines ingesting hundreds of external APIs under strict privacy and compliance controls.
    • Delivered HIPAA-aligned clinical workflow tools and interactive BI dashboards in React & TypeScript.
    PythonJavaC#/.NETNeo4jAirflowPresto/TrinoPostgreSQLReact

  3. + earlier engineering experience

Education

M.Eng, Computer Science

Dec 2024

University of Cincinnati

GPA 4.0

B.E., Computer Engineering

Savitribai Phule Pune University

03work

Selected work

Flagship case studies plus open-source tools. Client work is presented anonymized.

EvalsOSS
2026

upgrade-report

Personal project. A CLI that tells a team whether to switch models or prompts: per-example paired bootstrap, a measured noise floor, non-inferiority, and McNemar on pass/fail flips. Returns Ship on 97.2% of 500 simulated A/A runs.

PythonPaired bootstrapMcNemarpytest
Code Writeup
Infra
2026

inferno

Personal project. A from-scratch CPU inference engine in Rust for TinyLlama-1.1B. Against Ollama on CPU (Q4_0, 8 threads), prefill is about 1.3 to 1.6x faster and decode 1.2 to 1.35x. Prompt-lookup speculative decoding keeps greedy output identical.

RustQ4_0Speculative decodingTinyLlama-1.1B
Code Writeup
Agentic AIRAG
2025

Persistent Agentic Memory

Grounded LLM agents in domain data with pgvector and a Graphiti knowledge graph, so they recall prior interactions and cite their sources.

LangGraphpgvectorGraphitiPostgreSQL
Case study
Agentic AIEvals
2025

CI/CD & Eval Harness for Agents

Operationalized agent evaluation — gating task success, grounding, latency, and drift on every deploy with LangSmith.

LangSmithGitHub ActionsPythonDocker
Case study
RAG

Domain-Tuned Retrieval Embeddings

A proof of concept: fine-tuning bge-small on synthetic query-passage pairs lifted NDCG@10 by 4.59 points on 710 held-out queries.

bge-small-en-v1.5Qwen2.5-7BvLLMPython
Case study
Infra

Self-Serve Model Deployment

Upload a trained model through an admin panel and get a predict endpoint, with version and performance tracking, so ML engineers can ship classifiers without DevOps help.

FastAPIReactTypeScriptPostgres
Case study
Data

Operations Analytics for Manufacturing

Hourly exports into Snowflake and two dashboards built as code, so stakeholders on a manufacturing and warehouse platform can see effective labor across operators and stages.

Aurora PostgresS3KMSSnowflake
Case study
Data

Pre-Release Review of a Concurrency Feature

Found and ranked data-integrity risks in a multi-operator time-tracking feature before it was implemented.

TransactionsDB constraintsSchema validationAI-assisted review
Case study
Data

Impact-Tracing Knowledge Graph

A Neo4j graph linking IT assets to the business services they support, so a regulated enterprise can trace the business impact of an asset failure.

Neo4jFastAPIAngularag-Grid
Case study
Agentic AISecurity
2025

Jailbreak Defender

A jailbreak-defense sandbox: 17 regex patterns, a structural scanner, and a semantic firewall over 27 known attack vectors.

PythonLLM safety
Code
FrontendOSS
2024

NER Visualizer

Interactive tool for visualizing named-entity recognition output, built in TypeScript.

TypeScriptReactNLP
Code
04stack

Tools of the trade

The day-to-day stack, plus the broader toolkit I reach for across AI, backend, data, and frontend.

core stack
PythonTypeScriptLangGraphLangChainFastAPINext.js / ReactPostgreSQL / pgvectorAWSDocker

AI & Agents

LangGraphLangChainMCP workflowsRAG pipelinesMulti-agent orchestrationPersistent memory & contextLLM APIs (OpenAI, Anthropic)Prompt engineeringAgent evaluation (LangSmith)Responsible AI

Languages

PythonJavaTypeScriptJavaScript (ES6+)SQLC# / .NETBash

Backend & APIs

FastAPIFlaskREST APIsMicroservicesEvent-driven architectureKafkaNode.jsGraphQLWebSocketsServer-Sent Eventspytest / JUnit

Frontend & UI

ReactNext.jsReduxZustandJotaiData visualizationReal-time dashboardsViteesbuildTesting Library / Vitest

Data & Databases

PostgreSQLpgvectorNeo4j (Cypher)RedisMongoDBPresto / TrinoBigQuerySpark (PySpark)AirflowQuery optimization

Cloud & Infra

AWS (ECS, S3, Lambda, EMR)GCP (BigQuery, Dataproc)AzureDockerKubernetesTerraformGitHub ActionsGrafana / observability

Security & Compliance

HIPAA-aligned environmentsData governance frameworksAudit trailsAnti-jailbreak middlewareTenant isolation
daily ai tooling
Claude Code/Cursor/GitHub Copilot/research/prototyping/code review/Claude Code/Cursor/GitHub Copilot/research/prototyping/code review/Claude Code/Cursor/GitHub Copilot/research/prototyping/code review/Claude Code/Cursor/GitHub Copilot/research/prototyping/code review/

Currently building

updated October 2026
  • Owning the AI platform across client engagements: production RAG and agent systems, with LangSmith eval suites that gate merges in CI.
  • Shipped upgrade-report, a CLI that tells a team whether to switch models or prompts, with a writeup on how it decides (September 2026).
  • Shipped inferno, a from-scratch CPU inference engine in Rust for TinyLlama-1.1B, with a writeup on the build (September 2026).
Follow along on GitHub
06writing

Writing

Notes on agentic AI, evals, and shipping LLM systems in production.

Should we switch models? A tool that answers Ship or Don't ship

How upgrade-report uses paired statistics, a noise floor and non-inferiority testing to decide whether an LLM app should move to a new model or prompt.

The LLM Proposes, the Code Disposes: Building a Private Search Engine for My Photos

How I built natural-language search for my own photo library that runs entirely on one home machine, and why a pile of regexes beat a 14B LLM at understanding queries.

Writing an LLM inference engine from scratch in Rust

And making it faster than Ollama on my CPU

All posts
05contact

Let's build something agentic.

Hiring, collaborating, or just want to talk shop about agents and infra? My inbox is open.

Drop a note with the form — it reaches me directly, and I reply to everything.

open to relocation & remote across the US

find me