I'm Gaurav, an AI Engineer from UIC's MS CS program. I build local-first agents, RAG pipelines, and full-stack ML products. Recent work spans autonomous SRE remediation, synthetic data for tool-using agents, and a live inbox triage system you can try below.
Local-first AI inbox triage. A local scikit-learn + MiniLM classifier handles routing in ~5ms on CPU at $0. A PydanticAI agent runs only on uncertain cases. Nightly retraining loop learns from your corrections with regression guardrails.
Self-healing Kubernetes pipeline. A Go watchdog polls Prometheus, an MCP server bridges cluster infrastructure to an LLM, and CrewAI agents execute strategic merge patches. Escalates from restart to memory patch when the cheap fix fails.
Multi-agent synthetic data pipeline. Generates multi-turn tool-use conversations grounded in ToolBench schemas. LLM-as-judge scores each on naturalness, tool correctness, task completion, with automatic repair and diversity steering.
Production RAG API. Query academic docs in natural language at sub-200ms p95 with 95% retrieval consistency. Pinecone vector search, GPT-3.5-Turbo, local Flan-T5 fallback, GitHub Actions regression tests.
Semantic search across Pandas, NumPy, and Scikit-learn docs. An MLP classifier ranks retrieved snippets by source reliability, improving relevance by 20% by prioritizing official API refs over community content.
I'm an MS Computer Science student at the University of Illinois Chicago (May 2026, GPA 3.89), currently on OPT and looking for full-time AI Engineer, ML Engineer, and SDE roles at startups and mid-sized companies where I can ship product from day one.
My focus is agentic AI systems and RAG pipelines that work in production, not demos. Winnow runs on your laptop for $0 and only escalates to an LLM when needed. The SRE pipeline patches its own Kubernetes cluster. ToolGen generates the training data that makes tool-using agents actually good.
Before UIC I did an ML internship at AIML Analytics. As a Graduate Researcher at UIC I contributed to Pix2Path, a Transformer-based GAN pipeline processing 2TB+ of spatialomics data on the HPC cluster.
Redesigned the root-cause analysis prompt with structured multi-step reasoning (hypothesis generation, evidence validation, elimination). Added an alternative-hypothesis output section that reduced false-positive classifications by 30%.
I'm open to full-time AI Engineer, ML Engineer, and SDE roles. If you're hiring for agentic systems, RAG, or full-stack ML products, or you're a founder who needs the first ML hire, I'd love to hear from you.
Send me an email →