ISIshan
Ishan Srivastava, AI / ML Engineer

ISHAN

AI / ML Engineer

LLM System Design·RAG·Agentic AI·ML Pipelines

Open to a Fall 2026 AI/ML co-op.

0%

query deflection on enterprise chatbots

0%

faster ticket resolution

0%

manual effort cut via orchestration

0.0%

pipeline uptime

Ask About Me

Don't read a resume. Just ask.

A live agent grounded in my actual resume and projects. Tap a question, no typing needed.

Ask my portfoliogrounded

Pick one

About Me

Four years of production, then school on top of it.

I started at TCS in 2021 wiring analytics dashboards for a Life Sciences client, and spent four years working backwards from what broke in production. Dashboards became NLP pipelines. Pipelines became LLM chatbots running for three enterprise clients. Chatbots became orchestration with routing and automated decision points.

Now I'm doing a Master's in Applied AI at Northeastern on a 4.0, and building the thing I care about most: retrieval systems that stay grounded, agents that decide, and pipelines you can actually put a number on. I write my own eval sets because I'd rather know than guess.

I'm looking for a Fall 2026 AI/ML co-op in Boston. If your team is shipping LLM systems to real users, that's the work I want.

CURRENTLY

Master’s in Applied AI, Northeastern

Boston · 4.0 GPA · through May 2027

MOST RECENTLY

Tata Consultancy Services

Developer · Jul 2021 – Aug 2025

  • Deployed LLM chatbots across 3 enterprise clients (GPT-3.5/4) with intent classification, prompt engineering, and retrieval-grounded responses. 60% query deflection, +9 pt CSAT.
  • Built NLP pipelines for ticket classification, multi-turn FAQ with entity extraction, and summarization — 40% faster resolution at 99.5% uptime.
  • Led LLM orchestration with conditional routing and automated decision points. Cut manual effort 55%, lifted SLA compliance 88% → 96%.

  1. 2021

    Dashboards

    Started at TCS wiring real-time analytics (OSIsoft PI) for an enterprise Life Sciences client. Learned how production data actually behaves.

  2. 2022

    Python + NLP

    Grew into Python and NLP: ticket classification, entity extraction, document summarization. The first pipelines I owned end to end.

  3. 2023

    LLM Chatbots

    Deployed GPT-3.5/4 chatbots across 3 enterprise clients with intent classification and retrieval-grounded answers. 60% query deflection.

  4. 2024

    Orchestration

    Built LLM orchestration on internal tooling: conditional routing, automated decision points. Cut manual effort 55%; SLA 88% → 96%.

  5. 2025–26

    Agents & Eval

    Master's in Applied AI at Northeastern, 4.0 GPA. Now shipping eval-driven RAG and agentic systems with multi-stage guardrails.

Projects

Three systems, live and measured.

Each one is deployed and has an eval story. Use the arrows to move between them.

1 of 3

findmejob landing page: dark hero reading 'An AI career agent that actually knows you', with a private-beta badge

Findmejob (CareerForge)

Apr 2026

AI career platform · cost-aware multi-model routing

The problem

An AI career platform that assesses profiles, surfaces real jobs, and tailors a resume on click, without the per-user LLM bill spiraling.

Results

~$0.48
per-user/mo (vs ~$10 baseline)
70%+
prompt-cache hit target

How I built it

  • Cost-aware routing via Vercel AI Gateway: Sonnet 4.6 for moat tasks, GPT-4.1-mini for support, with prompt-caching targeting 70%+ hits.
  • Edit-via-JSON resume engine: the LLM emits structured edit ops, a deterministic transformer applies them to a stable LaTeX template (Tectonic in Vercel Sandbox), so no LaTeX-from-LLM bugs.
  • Supabase Auth (Google OAuth + email) with RLS; JSearch-backed job feed; rubric-grounded assessment.

Tech used

Next.jsVercel Fluid ComputeAI GatewaySupabaseTectonic
Live demo

Stack

What I reach for.

LLM System Design

RAG (chunking, retrieval, reranking)Agentic flows + function callingMulti-agent orchestrationMulti-stage guardrailsPrompt cachingCost-aware model routingToken & context-window management

Frameworks & Protocols

OpenAI APIClaude APILangChainLangGraphLangSmithVercel AI GatewayAnthropic MCPGoogle A2A

Retrieval & Embeddings

FAISSpgvectorsentence-transformersHugging FaceOpenAI embeddings

Evaluation & Reliability

RAGASGold-standard dataset designGroundedness scoringLatency optimizationMonitoringHard-refusal guardrails

Full-Stack

FastAPINext.js (App Router)ReactSSRServer actions

Cloud & Agent Hosting

AWS SageMakerGoogle ADKVercel (Fluid Compute, AI Gateway, Sandbox)RenderDigitalOceanSupabaseGCP

Infra & DevOps

DockerGitHub Actions CI/CDRedisPostgres

ML / DL & Tools

PyTorchscikit-learnspaCyNLTKPandasNumPyPythonClaude Code

Writing

Thinking in public.

Essays on where LLM systems actually break, and what the production data says.

All writing on Medium

Education

Where the fundamentals came from.

  • Northeastern University

    Master's in Applied Artificial Intelligence

    Sep 2025 – May 2027 (expected)
    GPA 4.0 / 4.0

  • IIIT Bangalore

    Advanced PG Certificate, Data Science

    Oct 2024 – May 2025

  • Mahatma Gandhi Kashi Vidyapith

    B.C.A., Computer Application

    2018 – 2021

certifications

  • Advanced Data Science & Machine Learning, IIIT-Bangalore (2025)Verify

Get In Touch

Ready when you are.

Open to Fall 2026 AI/ML co-op. Boston preferred, remote OK. If your team is shipping production RAG, agents, or LLM evals, I'd like to hear about it.