Open to opportunities

Keshav Singhal

@keshavsinghal

Message

AI-focused computer science student building evaluated LLM and real-time anomaly systems.

India

Message

What I'm looking for

I’m looking for a role where I can build and evaluate end-to-end AI systems—LLM/RAG pipelines, real-time anomaly detection, and practical APIs—while iterating from idea to measurable impact and working with a team that values solid evaluation.

I’m a CS undergrad with a strong interest in artificial intelligence and the systems built around it. I’ve worked across AI, from language models and retrieval systems to real-time data and anomaly detection, and I’m comfortable taking projects from idea to a working, evaluated system.

Recent work includes a Legal AI system using hybrid retrieval and a fine-tuned LLaMA model, with an evaluation framework pairing automated metrics and expert (lawyer) assessments (paper targeting JURIX 2026). I’ve also built a real-time threat detection pipeline with Apache Kafka and FastAPI using Isolation Forest/Autoencoders, plus LLM-based human-readable report generation, and I’m continuing to explore LLM-powered conversational simulation with controlled dialogue behavior and robust evaluation.

Experience

Work history, roles, and key accomplishments

Current

AI Medical Simulation Engine

Current

AI Medical Simulation Engine

Mar 2026 - Present (3 months)

Developed an LLM-powered clinical simulation engine supporting realistic multi-turn patient–doctor conversations. Built a FastAPI backend with session management and a hybrid evaluation approach combining rule-based scoring with LLM-generated qualitative feedback.

Multi Turn Conversational AI fastAPI Session Management Prompt Engineering

Current

JurisPro Legal AI System

Current

JurisPro

Built a hybrid RAG system for an Indian legal corpus combining BM25 sparse retrieval and FAISS dense retrieval. Fine-tuned LLaMA 3.1-8B on legal data and reduced training loss from 1.87 to 0.82 over 1500 steps, with an ablation-based evaluation framework aimed at a JURIX 2026 paper.

Retrieval Augmented Generation (RAG)LLM Fine tuning BM25 Faiss Reranking Model Evaluation Prompt Engineering

Current

Kafka Real-Time Streaming System

Current

Kafka-Based Real-Time Streaming System

Built a scalable Kafka ingestion pipeline using a producer-consumer architecture for continuous, high-throughput processing. Designed streaming workflows to support ongoing data ingestion and processing at scale.

Apache Kafka Producer Consumer Architecture Data Ingestion High Throughput Systems Scalability

Current

Real-Time Threat Detection

Current

AI-Powered Threat Detection & Analysis System

Created a high-throughput real-time network log pipeline using Kafka and FastAPI, generating interpretable anomaly reports via LLM integration. Implemented Isolation Forest and Autoencoder models for anomaly detection on CICIDS2017 and exposed monitoring and historical analysis via REST APIs.

Apache Kafka fastAPI REST APIs Anomaly Detection Isolation Forest LLM Integrations Network Log Processing Report Modeling

Education

Degrees, certifications, and relevant coursework

PES University

Bachelor of Technology, Computer Science and Engineering

2023 - 2027

Pursuing a B.Tech in Computer Science and Engineering at PES University, Bengaluru (expected 2027). Focused on building a strong foundation in AI and related systems.