Looking for a job

Sayantan Chaklader

@sayantanchaklader

AI application engineer building LLM-powered research, job-screening, and data-analysis apps with evidence-grounded answers and security checks.

India

Message

What I'm looking for

I’m transitioning from a biosciences background into AI application engineering, focusing on LLM workflows, agent orchestration, prompt engineering, testing, evaluation, security, and deployment.

My independent projects turn LLM workflows into practical applications, with reliability, evaluation, and security built into the design. I’m transitioning from a biosciences background into AI application engineering after learning Python and focusing on LLM systems.

For my Research Agent, I built a LangGraph workflow that searches the web, gathers evidence, and verifies support before trusting citations. I added SSRF-protected fetching and tested the system with a 2,100+ test Pytest suite.

My AI Job-Lead Pipeline screens AI jobs and sends a shortlist to Telegram. I evaluated its LLM judge on 55 labelled jobs, improving priority agreement from 58% to 70% with a prompt change, while catching all five injection attacks.

I also built an AI Data Analyst Agent that turns questions about uploaded CSVs into tables, charts, and explanations, with SQL validation and resource limits. My Executive PG Certification in Data Science & Artificial Intelligence is from IIT Roorkee iHUB DivyaSampark, with Intellipaat.

Experience

Work history, roles, and key accomplishments

RAG Document Q&A System Developer

Independent Project

Designed an end-to-end RAG pipeline covering document loading, chunking, embedding, vector storage and LLM-grounded generation with source-backed answers.

Python Lang Chain ChromaDB RAG Embeddings LLM

AI Job-Lead Pipeline Developer

Independent Project

Built a daily low-code n8n pipeline that collects AI jobs from two APIs, filters them with rules, has an LLM judge (Groq) return structured JSON with evidence quotes, and sends a shortlist to Telegram. Treated job text as untrusted input with two-layer prompt-injection defences that route to quarantine; added failure alerts, retries and HTML-safe messages after debugging failures in scheduled runs

N8n GROQ Google Sheets Telegram Docker Prompt Engineering LLM JSON API Integration

AI Data Analyst Agent Developer

Independent Project

Built an app where users upload a CSV, ask questions in plain English and receive tables, charts and AI-written explanations from LLM-generated DuckDB SQL. Treated generated SQL as untrusted: SQLGlot AST validation (read-only, single statement, table and function allowlists, result limits) runs before DuckDB executes anything, with DuckDB resource limits as a second layer.

Python GROQ DuckDB SQLGlot Streamlit Pytest SQL LLM Data Analysis

Research Agent Developer

Independent Project

Built a LangGraph workflow that plans sub-questions, searches, fetches sources, extracts evidence, critiques sufficiency and runs bounded follow-up research loops, with whole-run budgets on searches, fetches and LLM calls. Grounded every final claim in evidence with exact supporting quotes, then verified support semantically before any citation is trusted; answers that fail verification are reject