Founding engineer at Genrise.ai. Mathematics & Scientific Computing at IIT Kanpur.
I build infrastructure for AI agents. At Genrise that means a platform measuring how often brands get recommended across seven AI shopping assistants — which turns into cross-encoder fine-tuning, LangGraph agent harnesses, and MCP servers. In my own time it means shipping that same problem as open source.
Adding an agent to a Flutter app used to mean stitching together flutter_gemma, raw tool calling, your own ReAct loop, and your own memory store. This is all of that behind one API — and it runs fully on-device, with no network, if you want it to.
Seven providers through one interface: Gemini, OpenAI, Anthropic, HuggingFace, Ollama, on-device Gemma, and GGUF models. Five packages, four of them at a perfect 160/160 on pub.dev.
| Package | |
|---|---|
flutter_agentic |
Core runtime — ReAct loop, cost-aware routing, streaming, safety layer |
flutter_agentic_graph |
Stateful agent graphs — nodes, conditional edges, cycles, checkpointing |
flutter_agentic_memory |
Semantic memory — embeddings and an on-device vector store |
flutter_agentic_tools |
Device tools — clipboard, sandboxed files, sensors |
flutter_agentic_ui |
Drop-in chat UI — streaming bubbles, tool-call rendering |
genesis_ai is what I built on top of it: describe an agent in one sentence, get a working one running locally on Gemma 3N.
- Universal Memory Protocol — a probabilistic memory layer letting AI assistants carry context between apps that can't see each other. Recovers 94.5% of memory across disjoint clients at sub-4ms lookup. Preprint, 2026
- Impact of AI Usage on Critical Thinking Skills Among University Students — first author, JETIR 13(6), 2026
- Social Media Predisposition & Academic Wellness Among Secondary School Students — IJRAR 13(2), 2026
Traded Euribor and ESTR futures at Axxela, where I finished as the most profitable intern in my cohort and built an LSTM + XGBoost pipeline predicting daily SOFR fixes at R² 0.97. Earlier: formal verification work with DSPy and CBMC, LLM vulnerability matching for DRDO, and Flutter apps for the Government of India's Jal Jeevan Mission and the Government of Odisha.
Python · Dart · Go · PyTorch · LangGraph · DSPy · MCP · FastAPI · Postgres