AI Engineer / Forward Deployed Software Engineer

Open to AI engineering and YC startup roles

I build agentic systems that survive production.

Aditya R Rudra is a software engineer focused on RAG, AI agents, LLM workflows, and distributed data systems. He has shipped AI products at Wells Fargo, Blaze AI (YC W22), and Mati Labs across finance, legal, operations, and enterprise automation.

Talk about AI engineeringSee proof of workDownload resume

Case studies

Production AI work with real constraints.

The common thread is not model demos. It is turning ambiguous workflows into reliable systems with data pipelines, APIs, retrieval, observability, and user feedback.

Wells Fargo

Production AI over high-scale financial data

Software Engineer, Credit Cards & Accounts

Built RAG and agent-powered internal chatbots while supporting data services processing 2 TB/day with sub-150 ms P99 latency.

  • Enabled natural-language querying over structured financial data for internal teams.
  • Designed Kafka-backed microservices deployed on Azure for card-transaction workflows.
  • Built Flask, Django, Java, MongoDB, and React tooling used across engineering teams.
RAGAI agentsKafkaJavaPythonAzureMongoDB

Blaze AI (YC W22)

AI copilot for legal and enterprise workflows

Software Engineer

Led production copilot work with LangChain and BERT, cutting contract-analysis turnaround by 40%.

  • Integrated retrieval, NLP, and generation flows for contract analysis and document generation.
  • Built Spark batch jobs and GraphQL APIs serving thousands of AI pipeline queries per day.
  • Worked directly with founders and a fast-moving engineering team to ship customer-facing AI.
LangChainBERTSparkGraphQLPostgreSQLAWS EC2React

Mati Labs

Agent studio and client-specific automations

Forward deployed AI systems

Built and debugged multi-tenant agent workflows, connector credential flows, deployment paths, and client-facing AI automation.

  • Worked across agent orchestration, persistent memory, connector integrations, and runtime reliability.
  • Improved AI sales and operations workflows with CRM, S3 knowledge bases, vector retrieval, and email automation.
  • Operated close to deployment: local-first development, EC2 rollout, PM2 services, and production debugging.
LangGraphFastAPIQdrantS3NodeAWSPostgreSQL

Production client work

Three AI products shipped live for paying customers.

Real client engagements, real production traffic, real on-call. Each one was scoped, built, deployed, and handed off to a team that uses it every day.

01 · Client engagement

Live

AI coach that builds meal plans from blood work and lifestyle data.

Personalized nutrition coach, taken to production

Shipped a live AI coaching agent for a premium wellness brand — generates personalized food plans, runs a 6-pillar lifestyle analysis (Mindset, Nutrition, Sleep, Movement, Environment, Social), and grounds every recommendation in a knowledge graph + medical knowledge base.

  • Medical-report OCR pipeline over uploaded lab reports feeding a Qdrant-backed knowledge graph with citation-backed answers.
  • Calorie and macro targeting per client profile, with a Streamlit operator console for nutritionists to review and approve plans.
  • Live on AWS EC2 behind Nginx, CRM-integrated, monitored with Sentry and Prometheus, with a 480s streaming proxy for long generations.
FastAPIStreamlitAgnoQdrantKnowledge GraphMySQLRedisDockerAWS EC2Nginx

02 · Client engagement

Live

AI sales agent that writes, personalizes, and sends real customer email.

Outbound sales agent for premium programs, taken to production

Designed and shipped an outbound sales agent for a premium wellness brand — drafts, personalizes, and sends program-specific onboarding emails to inbound leads through the CRM, with a worker queue and full observability.

  • S3 knowledge base over programs, testimonials, and outcomes, retrieved via Qdrant for grounded, on-brand copy.
  • Program auto-detection from lead context, budget guardrails, follow-up logic, and brand signature enforcement to prevent personal-account leakage.
  • Production email pipeline on AWS EC2 with a worker queue, retry handling, and structured logging all the way out.
FastAPIAgnoQdrantS3Zoho CRMRabbitMQDockerAWS EC2

03 · Client engagement

Live

Control plane where specialized agents run per client, with approval gates.

Multi-tenant agent control plane, taken to production

Built and shipped a per-client, per-tenant agent control plane for a health & wellness operator — specialized agents (sales, operations, content, support) run against each client's data with persistent memory, approval gates, and per-company activity logs.

  • Express + Drizzle + PGlite backend with company-scoped data boundaries, atomic issue-checkout, budget hard-stops, and full activity logging.
  • LangGraph orchestrator routing between department-specific agent prompts, paired with a React + Vite board UI for operators to watch runs and approve gated actions.
  • Live on a dedicated EC2 (PM2-managed orchestrator + UI), in real use by client teams for per-tenant agent execution.
TypeScriptExpressReactViteDrizzlePGliteLangGraphPM2AWS EC2

Systems map

Strong in the middle layer between model and business.

YC and well-funded startups need engineers who can sit with users, understand the workflow, wire the data, choose the right model behavior, and ship the end-to-end product path.

Agent architecture

Tool-calling agents, retrieval, memory, guardrails, and job orchestration that map to real business workflows.

Data infrastructure

Kafka streams, Spark jobs, relational stores, document stores, and cloud services with production latency expectations.

Forward deployment

User interviews, workflow mapping, fast iteration, debugging in production contexts, and tight founder/customer loops.

Product engineering

APIs, internal tools, React interfaces, observability-minded services, and practical AI features that users can trust.

Map of a production AI agent system from data streams through retrieval, agent runtime, evaluation, and deployment.

Core stack

The tools I reach for first.

A short list, not a CV. These are the technologies I have used in production across multiple client engagements and full-time roles.

Python

FastAPI, Agno, LangChain, asyncio

TypeScript

React, Next.js, Node, Express

LangGraph

Multi-agent orchestration

AWS

EC2, S3, Lambda, SageMaker

Vector + KG

Qdrant, knowledge graphs, RAG

Data

Kafka, Spark, PostgreSQL, MySQL

LLM Ops

Prompt eval, guardrails, observability

DevOps

Docker, Nginx, PM2, CI/CD

Open source and research

Credibility beyond the job title.

GSoC contributor and mentor across two years, a peer-reviewed ns-3 workshop paper, and an active open source AI agent runtime.

Graduated from NITK Surathkal with a B.Tech in Computer Science and an 8.7/10 GPA.

WNS3 2023 — Linux-like socket statistics for ns-3

ACM Digital Library →

Published research on a Linux-style socket statistics utility for the ns-3 network simulator, presented at the Workshop on ns-3.

OpenMS — data-layout improvements (GSoC contributor 2021 with INCF)

GSoC archive →

Built data-layout and performance improvements for OpenMS, a C++ mass-spectrometry library, through Google Summer of Code with INCF.

GSoC mentor — OpenMS (2022)

GSoC archive →

Mentored a GSoC contributor on architecture and implementation for a bio-informatics data pipeline in OpenMS.

Available for AI engineering roles

Best fit: YC startups, AI-native teams, and forward-deployed engineering groups that need a builder who can own ambiguous systems end to end.

Email AdityaLinkedIn profileDownload resume