Tharanitharan M.
Resume

Chennai 13.08° N, 80.27° EBoston, Massachusetts 42.36° N, 71.06° W

TharanitharanMuthuthirumaran

Forward Deployed EngineerAI Engineer

I build AI software next to the people who'll use it. I sit in on the work, find where it gets stuck, then build the fix and train the team to run it.

Email meResume (PDF)GitHubLinkedIn
13 → 4 days
Request turnaround for 300+ caseworkers
Maine Needs
0
Incorrect AI-proposed database writes across 6 weeks of live use
Steady State
13%
Less OpenAI spend on Skillsoft's AI content platform
Skillsoft
324
NIST 800-53 controls searchable through an MCP server I published
AuditPilot

01The route

How I got here

My first two years out of school were consulting, working inside someone else's problem. Then I moved to product teams, where what shipped was mine to own. Forward deployed work turns out to need both.

  1. 2017 to 202313.08° N, 80.27° E

    Chennai

    Where it started

    I studied electrical and electronics engineering at Anna University, then joined Accenture. My client was Chubb North America, a regulated insurer, and the work rewarded patience: tuning SQL until their APIs ran 23% faster and tracing 150+ production defects back to the root cause.

  2. 2024 to 202543.66° N, 70.26° W

    Portland, Maine

    Consulting to product

    I came to the US for a master's in computer science at Northeastern's Roux Institute and finished with a 3.9. Product work started here. At Skillsoft I shipped LLM features and cut the platform's OpenAI spend 13%. At Maine Needs I worked on site with the team until I could see where requests piled up, then built what unstuck them. That took request turnaround from 13 days to 4.

  3. 2026 onward42.36° N, 71.06° W

    Boston, Massachusetts

    Where I am now

    At Steady State I built an AI backend that couldn't write to the database until a coach approved it. I'm based in Boston now.

02Work

Experience

  1. Jan 2026 – May 2026Boston, MAProduct

    Steady State LLC

    AI Engineer (Contract)

    • Built the backend and AI layer of an LLM coaching platform, with tool-calling and token-by-token streaming, that classifies athlete injury risk across 10 types and lets coaches chat with an AI about their athletes.
    • Required two-phase coach approval before any AI-proposed database write, with zero incorrect writes across 6 weeks of live use.
    • Tested the backend with Jest and Postman across unit, integration, API, security, performance, and load tests.
    • Wrote the system design, API contracts, and eval criteria that let 5 engineers build the rest of the app in parallel from week one.

    How the coach assistant works

    • TypeScript
    • NestJS
    • GraphQL
    • Node.js
    • Jest
    • Postman
  2. Sep 2025 – Dec 2025Portland, MEForward deployed

    Maine Needs

    Full Stack Engineer

    • Shortened request turnaround from 13 days to 4 for 300+ caseworkers by building and deploying an intake app with structured requests, live status tracking, and automated notifications.
    • Led on-site discovery with caseworkers to map the intake workflow, identify bottlenecks, and turn the findings into requirements for the intake app, then trained non-technical caseworker staff alongside the build.
    • Added an OCR pipeline on Google Cloud Vision that reads fields from uploaded documents, with rules-based validation and human review of low-confidence fields, ending manual spreadsheet entry for up to 600 requests a week at peak.
    • Next.js
    • React
    • TypeScript
    • PostgreSQL
    • Firebase
    • GCP
    • Cloud Vision
  3. Jan 2025 – Aug 2025Remote, USProduct

    Skillsoft

    Software Engineer (Co-Op)

    • Cut OpenAI spend 13% on the AI content platform through prompt and context trimming, feature toggles, and rollout guardrails, monitoring usage and LLM traces in Datadog to catch cost spikes early.
    • Launched an LLM auto-quiz generator in Author, Skillsoft's content platform, that writes publish-ready question sets for Codecademy and Percipio and saves 2 hours of manual authoring per course.
    • Shipped one-click localization on AWS (Amazon Translate), replacing an external vendor's manual XML translation process for French, German, and Spanish.
    • Migrated 6 production microservices from Python 3.12 to 3.13, standardizing deployment on a single Docker image and finishing the rollout with zero downtime.
    • Python
    • OpenAI
    • Datadog
    • Docker
    • AWS
  4. Oct 2021 – Dec 2023Chennai, IndiaConsulting

    Accenture

    Associate Software Engineer

    • Delivered 23% faster API response times for Chubb North America by tuning SQL, adding indexes, and reading execution plans.
    • Resolved 150+ production defects in Java and SQL services for this regulated insurance client through root-cause analysis.
    • Automated regression coverage in Pytest for an Agile team and wired it into Jenkins CI/CD, cutting manual QA cycles 19%.
    • Java
    • SQL
    • Pytest
    • Jenkins

Earlier · May 2021 – Jul 2021

Software Intern, Integration Wizards. Trained and validated computer vision models for custom object detection, and automated edge-device provisioning with Python, Bash, and Docker.

03Projects

Selected projects

Open source · Apache 2.0 · Multi-agent reference architecture

AuditPilot

A reference architecture for SOC 2 readiness reviews where no single model gets the last word. One agent drafts findings, a second agent tries to pull them apart, and a person signs off before anything ships.

Full case studyProject pageGitHubPyPInpm
How a finding moves through the system

  1. Evidence

    What needs checking against a control

  2. Orchestrator agent

    Drafts a finding, pulling controls through MCP

  3. Adversarial auditor agent

    Cross-examines each finding on its own

  4. Human approval gate

    LangGraph interrupt() holds the run until a person approves

  • compliance-kb-mcppgvector embeddings plus BM25 keyword ranking over 324 NIST 800-53 controls
  • Promptfoo + RAGASCI fails on any drop of more than 2% across 100 hand-labeled cases
  • Langfuse + OpenTelemetryEvery agent run traced, prompt versions pulled at runtime
5
MCP servers published
551
tests green at last full run
100
hand-labeled eval cases
16
architecture decision records
  • Published compliance-kb-mcp to PyPI and npm. It finds the right control among 324 NIST 800-53 controls with hybrid RAG retrieval, pairing pgvector embeddings with BM25 keyword ranking.
  • Gated GitHub Actions CI on a Promptfoo and RAGAS eval suite of 100 hand-labeled cases, so any change that drops agent quality more than 2% fails the build.
  • Traced every agent run in Langfuse and OpenTelemetry, with prompts versioned in Langfuse and pulled at runtime behind a local fallback.
  • Ran the build through a spec-driven Claude Code workflow: sub-agent reviewers for architecture, security, database, and compliance language, plus hooks that block destructive git commands.
  • Documented the major decisions in 16 ADRs so another team building human-gated, multi-agent compliance systems could adopt the pattern directly.

Stack

  • LangGraph
  • Pydantic AI
  • MCP
  • FastAPI
  • Next.js
  • Vercel AI SDK
  • Neon Postgres
  • pgvector
  • Promptfoo
  • RAGAS
  • Langfuse
  • OpenTelemetry
  • Cloud Run

  • AI coding mentor

    Mentivo

    Builds a personalized 7-milestone learning roadmap and checks understanding with adaptive quizzes.

    • Multi-turn mentor on Gemini 2.5 Flash through the Vercel AI SDK, with 8 versioned prompts and structured output validated by Zod.
    • Langfuse and OpenTelemetry trace latency, token usage, and prompt version across 16 API routes.
    • Promptfoo LLM-as-judge evals across 3 configs, plus 11 Vitest test files held to 80% coverage.
    LiveGitHub
    • Next.js
    • TypeScript
    • Gemini
    • Vercel AI SDK
    • PostgreSQL
    • Langfuse
    • Promptfoo
    • Zod

  • AI shopping assistant

    E-Commerce AI App

    A furniture storefront whose Gemini shopping assistant can search products, look up orders, and check out. Order history stays behind session authentication.

    • 70 unit, integration, and end-to-end tests with 90%+ coverage on the payment path. The suite runs in under 2 seconds.
    • Idempotent Stripe webhook with atomic multi-item inventory updates, and a per-user rate limit on the AI endpoint.
    • Production build scores 98 for Lighthouse performance, with sub-second page loads.
    LiveGitHub
    • Next.js
    • React
    • TypeScript
    • Gemini
    • Stripe
    • Clerk
    • Sanity
    • Vitest
    • Playwright

  • Full-stack banking app

    Horizon Banking Platform

    Links real bank accounts through the Plaid API for live balances and payment transfers.

    • API route protection, Zod input validation, and Sentry error tracking on every authenticated path.
    • Chart.js dashboards for spending patterns and transaction history.
    GitHub
    • Next.js 14
    • TypeScript
    • Tailwind CSS
    • Plaid API
    • Appwrite
    • Chart.js
    • Sentry
    • Zod

Also built

04Toolkit

Skills, and where I used them

Each group lists the jobs and projects where that work happened, so you can check it against the sections above.

  • AI engineering

    • LLM applications
    • Agentic workflows
    • Multi-agent systems
    • LangGraph
    • MCP
    • Tool-calling
    • RAG
    • Pydantic AI
    • Vercel AI SDK
    • OpenAI, Anthropic, Gemini APIs
    • Prompt engineering
    • Claude Code

    Used at Steady State · Skillsoft · AuditPilot · Mentivo

  • Evals and observability

    • Promptfoo
    • RAGAS
    • LLM-as-judge
    • Guardrails
    • Langfuse
    • OpenTelemetry
    • Datadog
    • Grafana
    • Sentry
    • PostHog

    Used at AuditPilot · Mentivo · Skillsoft

  • Full-stack

    • TypeScript
    • Python
    • JavaScript
    • Java
    • SQL
    • Bash
    • React
    • Next.js
    • Tailwind CSS
    • Node.js
    • NestJS
    • GraphQL
    • FastAPI
    • Spring Boot
    • REST APIs

    Used at Maine Needs · Steady State · E-Commerce AI App · NBA Analytics

  • Data

    • PostgreSQL
    • pgvector
    • Neon
    • Firebase
    • MySQL
    • Redis
    • DynamoDB
    • Kafka

    Used at Maine Needs · AuditPilot · Stock Market Pipeline

  • Cloud and DevOps

    • AWS
    • GCP
    • Azure
    • Docker
    • Terraform
    • GitHub Actions
    • Jenkins
    • Git

    Used at Skillsoft · Maine Needs · AuditPilot · Accenture

  • Testing

    • Jest
    • Vitest
    • Pytest
    • Playwright
    • Postman

    Used at Steady State · Accenture · E-Commerce AI App

05Credentials

Certifications and coursework

Claude Academy by AnthropicCourse completions

6 Claude Academy courses, completed

Anthropic's courses on building with Claude, the Model Context Protocol, and working alongside agents, finished in September 2026. Each pair notes where that material shows up in my own work.

  • Building on Claude

    Where it shows upAuditPilot was built through a spec-driven Claude Code workflow with sub-agent reviewers and guard hooks.

  • Model Context Protocol

    • Introduction to Model Context ProtocolCompleted Sep 17, 2026 · Verify badge
    • Model Context Protocol: Advanced TopicsCompleted Sep 17, 2026 · Verify badge

    Where it shows upAuditPilot ships 5 MCP servers, including compliance-kb-mcp on PyPI and npm.

  • People and agents

    • AI Fluency: Framework and FoundationsCompleted Sep 16, 2026 · Verify badge
    • Building Effective Human-Agent Teams (beta)Completed Sep 16, 2026 · Verify badge

    Where it shows upHuman approval gates decide what an agent may change, at Steady State and in AuditPilot.

Proctored certifications

  • AWS Certified Cloud Practitioner (CLF-C02)

    Amazon Web ServicesIssued Dec 2025Valid through Dec 2028

    Verify

  • Microsoft Azure Fundamentals (AZ-900)

    MicrosoftIssued Aug 2024Credential ID 1FF779DC01930B8E

    Verify

06Education

Two degrees, two countries

  • Northeastern University · Roux Institute

    Master of Science in Computer Science

    GPA 3.9 / 4.0Portland, MEJan 2024 – May 2026

  • Anna University

    Bachelor of Engineering in Electrical and Electronics

    GPA 3.6 / 4.0Chennai, IndiaAug 2017 – May 2021

Outside work

  • 8x hackathon winner

    Including first place at the Roux Hackathon, where I led a 5-person team to an AI-driven MVP and a $2,000 prize.

  • VP, Startup Club at Northeastern

    Ran 4 founder-focused events with about 40 attendees each.

07Contact

Get in touch

I'm looking for full stack engineer or AI engineer roles, and I'm open to relocating anywhere in the US. I have valid work authorization and don't need sponsorship right now. If you have a team that needs someone to sit with users and ship the thing, I'd like to hear about it.

muthuthirumaran.t@gmail.com
LinkedInGitHubResume (PDF)

Ask about my work

It runs on Gemini, answers only from the facts on this page, and keeps replies short.

Hi, I'm an assistant that answers from the same facts as this page. Ask me about Tharanitharan's work, projects, or the roles he's after.