Open to work · AI Product Manager

I build AI productsthat actually work.

AI Product Manager & Builder. I ship ML systems for fraud detection, risk intelligence, and LLM safety from model to API to production.

Explore RiskOS View GitHub

AI Product Manager & Builder · Hyderabad, India

Souptik Chakraborty

AI Product Manager & Builder · Hyderabad, India

I sit at the intersection of product thinking and ML engineering. I don't just spec features - I build them. RiskOS is my flagship: a 4-service open-source risk intelligence platform running live on HuggingFace with real APIs you can call right now.

Product Strategy

ML Engineering

Fraud & Risk

LLM Systems

API Design

souptikchakraborty

GitHub

Souptik96 · 5 repos

HuggingFace

soupstick · 7 spaces

FLAGSHIP PROJECT

RiskOS

Open-source AI risk intelligence platform. 4 services. All live. All callable via API.

Open Source

MIT License

AI Risk Intelligence Suite

Four composable services for fraud scoring, transaction triage, LLM safety, and marketplace analytics. Running on HuggingFace Spaces. No signup required. Call the API right now.

≥88%

Fraud Recall

~70%

Triage Reduction

~94%

LLM Block Rate

55ms

Inference Latency

🔍

Fraud Intelligence

XGBoost + SHAP

5-agent suite: transaction fraud, credit risk, KYC identity, sanctions screening, and LLM risk consultant.

Automated triage engine. Auto-closes safe transactions. Escalates only what needs a human eye.

ESCALATE

🛡️

LLM Guard

LangChain + RAG + Opik

RAG-augmented guardrail layer. Blocks jailbreaks, prompt injections, and PII leakage. Every call audited via Opik.

BLOCKED

📊

Marketplace Intelligence

NL -> SQL -> Plotly

Ask questions about marketplace data in plain English. SELECT-only security. Auto chart generation.

QUERY OK

Live Demo GitHub

User -> FastAPI Gateway -> Fraud (XGBoost) | Pipeline (LightGBM) | Guard (RAG) | Marketplace (NL -> SQL) -> Structured JSON

TRY IT LIVE

Real API calls. Real responses.

These hit live HuggingFace Spaces. First call may take 20-30s on cold start.

💡 HuggingFace free tier spaces sleep after inactivity. If your first call takes 30+ seconds, that's normal. The space is waking up. Subsequent calls are fast.

Fraud Intelligence

5-agent suite: transaction fraud, credit risk, KYC identity, sanctions screening, and LLM risk consultant.

/api/v1/fraud/predict

Editable JSON payload

Waiting for response...

ALL PROJECTS

What I've built

ML experiments, fine-tuned models, and live tools.

🔐

RiskOS

4-service AI risk intelligence platform. Live APIs, open source, MIT licensed.

XGBoost

LightGBM

FastAPI

Docker

GitHub Live

🧠

SmolLM3 QLoRA

Fine-tuned SmolLM3-3B on OpenOrca + SlimOrca for instruction following with RAG.

QLoRA

Transformers

SmolLM3

RAG

GitHub Live

🛒

Qwen2-VL Amazon LoRA

Multimodal LoRA adapter. Input product image -> output Amazon listing copy.

Qwen2-VL

LoRA

Multimodal

E-commerce

GitHub Live

🔎

SearchQual

Search quality evaluation tool built on Streamlit + Docker. Live on HuggingFace.

Streamlit

Docker

Evaluation

GitHub Live

OPEN TO WORK

What I'm Looking For

I'm looking for an AI Product Manager role where I can turn early ideas into shipped AI systems with real users, metrics, and product judgment behind them.

->Own 0-to-1 AI product development

->Work at the intersection of ML engineering and product strategy

->Build in fintech, compliance, or developer tools

I bring an unusual combination: product judgment + the ability to ship ML systems using AI coding agents without an engineering team.

Book a 30-minute call Email me directly

Let's talk.

Building something in fraud, risk, or LLM safety? Want to collaborate or say hi?

LinkedIn: souptikchakraborty