Founding AI Research Engineer @ Birla AI Lab (Aditya Birla Group)
Building India-first, production-grade AI systems — from Time-Series Foundation Models to LLM-native apps that actually ship.

• Bharat TSFM — spearheading India’s first Time-Series Foundation Model for industrial forecasting & decisioning.
• Birla Forecasting App — Next.js + FastAPI + Supabase + Docker on AWS (EC2 + Nginx), unified datasets → forecasts → chat.
• TSFM R&D — Chronos fine-tuning (CPI/soil-moisture), TimesFM zero-shot pipelines (WPI), LoRA/QLoRA, patchification, RevIN/DAIN.
• Curiosity Block (EdTech) — AI-native learning tools (targeting BITS Pilani & Birla schools cohort).
• Voice/Multimodal — Sarvam Voice Agent (multilingual), audio segmentation, Gemini→Stable Diffusion prompt bridge.
• Infra — Zephyr-7B Medusa FastAPI server (async), uv/uvx packaging, CI-ready CLIs.

• Time-Series AI: Foundation models, zero-shot baselines, robust scaling, quantile heads, evaluation harnesses.
• LLM Apps: Retrieval & tool-use, multilingual UX, latency-aware inference, productizing research.
• ML Infra: Dockerized services, AWS (EC2/Nginx), FastAPI backends, dataset registries, telemetry, and observability.

• Languages: Python • Rust • TypeScript • CUDA (C++ & Triton)
• DL/Serving: PyTorch • JAX • TensorFlow • Uvicorn • FastAPI
• Time-Series: Chronos • TimesFM • PatchTST • N-BEATS/N-HiTS
• LLM Ops: LoRA/QLoRA • PEFT • RLHF • SFT • Prompting • RAG
• Data/Infra: Next.js • Supabase • Docker • AWS EC2 • Nginx • uv/uvx packaging
• Extras: LangChain • Vector DBs • ngrok (dev tunnels)

• Bharat TSFM — Industrial-grade Indian TS foundation model; focus on scale, reliability, and business KPIs. (R&D; enterprise)
• Birla Forecasting App — End-to-end forecasting platform (Next.js + FastAPI + Supabase + Docker on AWS); ingest → forecast → explain → chat. (shipping)
• TSFM Evaluation Suite — Zero-shot TimesFM pipelines for WPI; Chronos fine-tuning on CPI & soil moisture; quantile evaluation; robust scaling. (research + code)
• Curiosity Block — AI-native learning tools for students; emphasis on curiosity over copy-paste; early pilots with academic partners. (edtech)
• Sarvam Voice Agent — Multilingual voice assistant + audio segmentation; real-time pipelines and TTS/ASR loops. (multimodal)
• Zephyr-7B Medusa Server — Async FastAPI server (8-bit), queueing & ngrok integration for fast dev loops. (infra)

• Production over prototypes: every demo has a path to deploy.
• Latency & reliability first: measure p50/p95 from day one.
• Human-in-the-loop: design interfaces that improve operator judgment.
• Curiosity > copy: products should teach and amplify.

⭐️ If something here helps you, star a repo or say hi — always open to collabs.