Kevin Kakolla

Senior AI Engineer — ML · GenAI · Agentic systems

GitHub HuggingFace YouTube LinkedIn X CV Consulting

About

I build production AI systems that reach real users: agentic workflows, RAG pipelines, tool-calling, LLM evaluation and guardrails, and fine-tuning (QLoRA/DPO). 15+ years of software engineering, the last 3+ shipping ML/GenAI in production, with 20+ enterprise customers delivered across stacks and domains. Currently completing a Graduate Certificate in Generative AI & LLMs at Carnegie Mellon University (2026).

I build in public — code on GitHub, models on HuggingFace, and build walkthroughs on my YouTube channel Thinking Machines.

Consulting

I take fixed-scope AI engagements — voice agent MVPs, RAG pipelines with evals, and agent audits — shipped in 1–2 weeks. See offers & contact →

Selected projects

Job Application Agent An AI agent that discovers roles, tailors resumes, and fills applications while I review and hit submit — deliberately human-in-the-loop. Full build video.
Voice Agent "Chitti" Real-time voice AI agent — Pipecat · Twilio · Deepgram · Cartesia · GPT-4.1 — running at sub-1s latency in production.
Mind2Web Computer-Use Agent Two-stage DeBERTa web-agent grounding model; 93% recall@10 on unseen domains. Published on HuggingFace.
Sales Voice Agent Specialized LLM powering next-generation sales conversations for an energy company.

Publications & releases

mind2web-candidate-ranker (HuggingFace, 2026) — fine-tuned DeBERTa-v3 grounding model for web agents: ranks candidate DOM elements for a given action; 93% recall@10 on unseen domains (Mind2Web benchmark). Training & eval code.

News

Jun 2026 — Published “My AI Agent Filled Job Apps”, a 43-minute build walkthrough of the job-application agent.
2026 — Completing CMU’s Graduate Certificate in Generative AI & LLMs.
2025–26 — Shipped production voice agents and enterprise RAG/agentic pipelines.

Contact

Email: kkakolla@andrew.cmu.edu — or DM on X.