Full-stack AI Engineer in Bengaluru (Bangalore), India. I build AI agents, RAG pipelines and LLM applications — plus the backends and interfaces around them. Currently AI Workflow Engineer at DemandNXT.
Most of my work lives in the gap between "the model gave an answer" and "the system did the right thing reliably." Orchestration, structured outputs, validation, and handling the ways models fail.
| Project | What it is |
|---|---|
| Maestro · live | Multi-model orchestration engine. A conductor routes one task across thinker / worker / verifier roles and returns a verified answer with a full, replayable decision-log. |
| Cannon · live | Multi-agent assistant. Each domain is an independent specialist — isolation enforced at the query, not by convention. Typed provider fallback (Groq→Gemini). 91 unit + 13 e2e tests. |
| ReplyDesk | WhatsApp lead agent replying in ~8s, with a live dashboard for lead feed, pipeline and response-time KPIs. |
| CraftConnect | Multi-modal assistant letting artisans run a storefront by talking and showing. Semi-finalist, Google Cloud Gen AI Exchange Hackathon 2025 (270,000+ developers). |
| AI Notes | Notes app running DeepSeek R1 locally via Ollama — inference never leaves the machine. |
AI-Powered Note-Taking System: A Local Machine Learning Approach DeepSeek R1 Integration — IJRTMR, Nov–Dec 2025, pp. 178–189. DOI: 10.59256/ijrtmr.20250506023
Python · TypeScript · React · Node · FastAPI · Postgres · Gemini · Groq · Llama · DeepSeek · Ollama · Google Cloud · n8n
Portfolio · LinkedIn · harshith28124@gmail.com · harshith29124@gmail.com