Skip to content
View Vk2245's full-sized avatar

Highlights

  • Pro

Block or report Vk2245

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
Vk2245/README.md
Vishal Kumar - AI Engineer, Agentic AI, RAG, Production ML

Typing SVG

Vishal Kumar

AI engineer building production-oriented AI systems across agents, RAG, Production ML, streaming inference, and LLM evaluation.

I am a B.Tech CSE student at IIIT Nagpur, graduating in 2027. My work is focused on systems that can be evaluated, deployed, and inspected: multi-agent workflows, grounded retrieval, fine-tuning pipelines, safety guardrails, and low-latency ML inference.

Current focus: agentic AI, multi-agent systems, RAG, reinforcement learning, LLM evaluation, production ML inference.

Proof

Signal Result What it shows
AMD Slingshot AI Promptathon #1 / 823 Prompting, evaluation, fast system design
HackerRank Orchestrate May 2026 Top 1.2% (#159 / 12,885) Grounded agent workflows across 48 countries
Wunder Fund LOB Predictorum Top 2.4% (#120 / 4,917) Streaming ML, quantitative modeling, inference optimization
ArtPark CodeForge, IISc Bangalore 1st runner-up Hackathon execution and product thinking
Triagegeist Clinical AI OOF QWK 0.9987 Clinical feature engineering and model validation

Selected Systems

Project Scope Stack Links
CrisisOps Multi-agent SRE training environment with procedural incidents, tool-based agents, automated judging, and RL-style reward design. Python, LangGraph, OpenEnv, GRPO, QLoRA, FastAPI, Hugging Face Repo, Live
Triagegeist Clinical triage prediction system with hierarchical safety logic, 200+ engineered clinical features, and gradient-boosted ensemble modeling. XGBoost, LightGBM, CatBoost, Production ML, feature engineering Repo
LOB Predictorum Streaming-native limit order book prediction system built after diagnosing batch-to-live degradation. PyTorch, BiGRU, TimeMixer, InfoNCE SSL, ONNX, INT8 quantization Repo
WasteWatch AI AI-native civic reporting platform with vision classification, model fallback, API integrations, and bilingual voice workflows. Gemini, Groq, FastAPI, React, voice AI Repo, Live
CivicPath Non-partisan election assistant for Indian voters using multilingual guidance, fact checking, and Google services. Gemini, Google Cloud Run, Maps, Speech, multilingual NLP Repo, Live
DevBoost AI Agentic code analysis tool with explain, test, and fix modes, structured output contracts, and streaming responses. React, Vite, FastAPI, LangChain, IBM Bot API Repo

Engineering Areas

AI systems: agent orchestration, tool use, RAG, prompt contracts, hallucination controls, LLM evaluation, fine-tuning.

ML and research: Production ML, time-series modeling, ranking metrics, self-supervised learning, model compression, ablation-driven iteration.

Production: FastAPI services, React interfaces, Dockerized apps, cloud deployment, API integrations, streaming inference, observability-minded design.

Research Interests

  • Reliable multi-agent systems with measurable task success.
  • RAG systems that cite evidence and fail safely when evidence is weak.
  • Reinforcement learning and reward design for tool-using agents.
  • Clinical and civic AI systems where false confidence is more dangerous than abstention.
  • Low-latency inference for streaming and decision-support workloads.

Contact

I am open to AI engineering internships, research engineering work, and founder-led teams building serious applied AI systems.

Pinned Loading

  1. CrisisOps-Multi-Agent-SRE-Training-via-OpenEnv CrisisOps-Multi-Agent-SRE-Training-via-OpenEnv Public

    A procedural reinforcement learning environment simulating cascading microservice failures. Trains LLM agents (Qwen3-8B) as cooperative Site Reliability Engineers using a "Buddy System," Difference…

    Python

  2. Triagegeist-vk2245 Triagegeist-vk2245 Public

    Three-tier hierarchical clinical AI for ESI triage prediction | OOF QWK 0.9987 | LightGBM + CatBoost + XGBoost ensemble with NEWS2, CCI, Modified Shock Index | Triagegeist Competition 2026

    Jupyter Notebook

  3. -Vk224-Wunder-Challenge-LOB-Predictorium -Vk224-Wunder-Challenge-LOB-Predictorium Public

    LOB Prediction Challenge: Rank 120/4917 (Top 2.4%) | Failed with 17 tree models (0.006), pivoted to BiGRU+TimeMixer ensemble with INT8 quantization (0.2812). Complete journey from batch processing …

    Python

  4. WasteWatchAI WasteWatchAI Public

    HTML

  5. DevBoost-AI-IBM-Bob-Code-Analysis-Platform DevBoost-AI-IBM-Bob-Code-Analysis-Platform Public

    AI-powered code analysis tool built with React + Vite for IBM Bob Dev Day Hackathon. Features three analysis modes: Explain (detailed code breakdowns), Tests (auto-generate test suites), and Fix (s…

    JavaScript

  6. orchestrate-triage-agent orchestrate-triage-agent Public

    Terminal-based multi-domain support triage agent for HackerRank Orchestrate. Uses only the provided corpus to classify tickets, retrieve evidence, decide reply vs escalate, and generate grounded re…

    Python