AI systems engineered for <production_scale>
I'm Chetan — an AI/ML Engineer building production LLM and RAG systems for fintech and risk intelligence platforms. I fine-tune transformer models with LoRA, wire automated RAGAS evals & guardrails into pipelines, and deploy async FastAPI microservices that hold response latencies sub-second and uptime at 99%+.
Production Architectures & Evals
Not prompt wrappers — real evaluation harnesses, adapter pipelines, and consensus systems shipped in fintech and risk domains.
Risk Minor
PRODUCTIONMulti-agent risk intelligence platform for regulatory filings. Combines BM25 keyword search with Milvus dense vector retrieval, a fine-tuned Llama 3.1 8B inference layer, input/output validation guardrails, hallucination-detection fallbacks, and real-time RAGAS + Langfuse quality tracing across all pipeline nodes.
Multilingual LoRA Fine-Tuning
PRODUCTIONFine-tuned Llama 3.1 8B using LoRA with adapter merging across 8 Indian languages — Assamese, Bengali, Gujarati, Kannada, Marathi, Nepali, Tamil, Malayalam — enabling automated legal document translation and multilingual regulatory compliance support with preserved language fidelity.
Fact Check Multi-Agent
OPEN-SOURCEFake-news detection engine running 3 local LLMs — Qwen 2, Phi-3, LLaMA 3.1 — in parallel via Ollama, aggregating verdicts through a weighted consensus layer with optional DuckDuckGo live web context injection to eliminate single-model hallucination and recency bias.
Live RAG Architecture Simulator
Experience how queries travel through Chetan's production pipeline — from dense/sparse retrieval and LoRA inference to RAGAS hallucination guardrails.