Skip to content
Rishabh Hota

Services

Building, training, consulting: intelligence in the flow of work.

A short call fixes the problem, the data and the done-condition. A working prototype lands within days, not a deck, and the shipped build comes with an eval suite, documentation and a handover, so it keeps working after he leaves.
  1. 01

    build · agents and automation

    Agentic systems that replace a process

    Multi-agent workflows with tool access, guardrails and evals, built to take over a whole process rather than a single step.

    Multi-agent systems, agentic RAG workflows, Model Context Protocol and agent-to-agent collaboration, brought into the tools a team already uses rather than a demo beside them, with review and approval steps kept in the loop. At Simplilearn the Collaterals Agent cut course-launch collateral from 7–10 days across two teams to one day, and won the Best Team Award.

    LangGraphGoogle ADKCrewAIn8nMCPA2Aevals
    Email about this ↗
  2. 02

    build · rag and knowledge

    Retrieval systems that are measured, not assumed

    Production RAG and Graph RAG over your documents: hybrid search, re-ranking, and an eval suite that runs in CI so quality is a number.

    Hybrid retrieval (FAISS and BM25 fused by reciprocal rank, then cross-encoder re-ranking), a free-model fallback chain and a security gateway. Built sales-enablement platforms on agentic RAG that generate learning content, scripts and playbooks adapting in real time. The agent on his own site runs this way at zero hosting cost: 35 retrieval, fact and guardrail checks passing, RAGAS faithfulness 0.97.

    RAGGraph RAGFAISSBM25RAGASFastAPI
    Email about this ↗
  3. 03

    build · fine-tuning and serving

    Small models tuned for one job, served on your hardware

    LoRA and QLoRA fine-tunes of Gemma, Llama and Phi with Unsloth and TRL, then quantised and served with vLLM, Ollama or llama.cpp, down to on-device.

    Supervised fine-tuning on your data in 4-bit and 8-bit, KV-cache and batching work for throughput, OpenAI-compatible serving. Ran Gemma 3n offline on Android inside a custom chatbot app, and long-form speech generation with Qwen 3 TTS: thirty minutes of consistent narration without drift.

    UnslothLoRA · QLoRATRLvLLMllama.cppspeech · TTS
    Email about this ↗
  4. 04

    build · medical imaging and vision

    Clinical imaging pipelines, from DICOM to report

    X-ray and CT classification, segmentation and reporting on AWS, with doctor-validated feedback loops for retraining.

    Led clinicians and engineers at OsteosConnect to a 96%-accurate osteoarthritis pipeline (X-ray to S3, Lambda, SageMaker and the EMR) with over 90% concordance with radiologists in hospital trials, ahead of the competitor RBKnee. LCSNet, his lightweight caries segmentation network, is published in Elsevier's Intelligence-Based Medicine.

    computer visionDICOMSageMakersegmentationAWS
    Email about this ↗
  5. 05

    build · applied research

    New research, turned into a working demo

    A paper or a release on Monday, a code-backed prototype by Friday: MCP, agentic patterns, new model families, efficient attention.

    Reading papers, forming bold hypotheses, testing in code, then hardening and scaling what works with robust MLOps and LLMOps. At Simplilearn this is the day job: dissect emerging AI into hands-on prototypes, document them, and feed the roadmap. Designed QFPViT, a quantum self-attention operator for Vision Transformers, starting from zero quantum background; the paper is with Quantum Journal.

    applied researchprototypingMLOpsLLMOpsquantum attentionpublications
    Email about this ↗
  6. 06

    teach · training and workshops

    Teams that can build agents, not just discuss them

    Code-driven modules and learning hubs for engineering and product teams: agents, RAG, evaluation and MCP, taught from working repositories.

    Bite-sized modules with runnable code, a curated hub teams come back to, and evaluation habits that outlast the session. This is how the research and learning modules at Simplilearn are built.

    workshopscurriculumagentsRAGevals
    Email about this ↗
  7. 07

    consult · ai strategy

    AI strategy that ends in a deployed system

    Consulting for teams that want to design the trend rather than follow it: which model, which architecture, what to build first, and how to run it in production.

    An AI researcher, engineer and consultant strategist who helps businesses not just follow trends but design and deploy them in production. GenAI tools and copilots designed and shipped in ed-tech and finance, bridging prototypes to production; breakthroughs turned into enterprise-ready systems that deliver measurable intelligence in the flow of work.

    AI strategycopilotsarchitecturemodel selectionproduction
    Email about this ↗