Ashburn, VA · Healthcare & Clinical AI

Bhaskar Gurram

AI Engineer — Agentic Systems · Multimodal RAG · Biomedical AI. Building production RAG at 50M-embedding scale, multi-agent clinical research systems, and EEG–fMRI neural decoding models.

GitHubScholarLinkedInEmailResume

01Selected work

Systems that hold up in production — and in peer review.

Open source · Agent safety · MCP

01

Unwind

A reversibility layer for agentic tool use. A transparent MCP proxy that works out which agent actions can be taken back, quietly reverses the ones that go wrong, and interrupts you only for the ones that genuinely can't be undone.

  • Human oversight of agents fails because approval prompts are undifferentiated — nothing in the stack knows which actions are reversible, so people develop an approve-approve-approve reflex. Unwind supplies the missing primitive.
  • Contributed the R0–R4 reversibility taxonomy, treating reversibility as ordinal and environment-relative: the same write_file is self-reversible on a git-backed tree and irreversible on a versionless one.
  • Fails safe asymmetrically — misclassifying an irreversible action as reversible is catastrophic while the reverse is merely annoying, so unknown tools, failed classification, and timeouts all escalate to a human.
  • Apache-2.0 with full open-source governance: CODEOWNERS, security policy, code of conduct, citation metadata, OpenSSF Scorecard, CI with coverage, and a published docs site.
R0–R4
reversibility classes
3
distribution channels
9
MCP clients supported
Apache-2.0
license
↗ source↗ docspip install unwind-mcp
PythonTypeScriptMCPTyperDockeruvpytestGitHub Actionsmkdocs

Agentic AI · RLHF

02

Autonomous Multi-Agent Research Platform

A Planner–Retriever–Reasoner–Validator agent loop with persistent memory and tool use across PubMed, ArXiv, and Semantic Scholar APIs.

  • PPO reward model trained on 12K researcher preference pairs — +41% synthesis quality and −68% hallucinated citations vs base GPT-4o on SciEval.
  • Deployed on GKE with LangSmith, Prometheus, and Grafana observability.
+41%
synthesis quality
−68%
hallucinated citations
5K+
concurrent sessions
<200ms
median latency
PythonLangGraphGPT-4oPPOFAISSNeo4jFastAPIPostgreSQLRedisKubernetes

Clinical AI · Knowledge Graphs

03

Multimodal Clinical AI Assistant

GPT-4V vision, FHIR parsing, Neo4j patient graphs, and vector search combined for knowledge-graph patient profiling.

  • Longitudinal biomarker-trend layer via graph traversal — 87% anomaly-detection sensitivity, 16 points above rule-based baselines.
  • HIPAA-aligned microservices on AWS EKS; adopted by 3 pilot healthcare teams.
94%
retrieval accuracy
15+
biomarker categories
87%
anomaly sensitivity
3
pilot healthcare teams
PythonGPT-4VLangChainFHIRNeo4jChromaDBStreamlitFastAPIDockerAWS EKS

Real-Time Retrieval · Source Verification

04

Conversational AI Search Engine

QLoRA-fine-tuned Llama 3.1 with hybrid dense + sparse retrieval and automated citation and fact-verification.

  • −40% misinformation on TruthfulQA-adapted benchmarks; provenance tracking with confidence scoring and claim decomposition (+35% NPS).
  • vLLM streaming serving 1K+ concurrent users on 4×A10G GPUs.
−40%
misinformation
150ms
P95 time-to-first-token
1K+
concurrent users
+35%
NPS
PythonLlama 3.1QLoRAKafkaElasticsearchQdrantFastAPIReactKubernetes

M.S. Thesis · Biomedical AI

05

EEG–fMRI Fusion for Neural Decoding

A multimodal neuroimaging framework predicting visual stimuli by fusing 70-channel EEG with 3T fMRI (Wakeman–Henson dataset) using temporal convolutional networks.

  • SSS, ICA/wavelet-ICA, and DWT preprocessing; MNI-normalized fusiform/occipital ROIs.
  • Beat GRU, LSTM-CNN, and SVM baselines — vs 65.5% EEG-only and 74.6% fMRI-only.
84.8%
within-subject accuracy
81.1%
cross-subject (LOSO)
0.93
ROC-AUC (within)
0.90
ROC-AUC (cross)
PythonPyTorchMNENilearnTCNs

02Research

Agent evaluation, selective risk, and multimodal neural decoding.

Preprints — 2026

Journal papers — 2021

  • 2021

    Gurram, B. A Review on Secure Data Transmission for Banking Application using Machine Learning. With Reddy, M.D. & Thatikonda, M.

    IJEAT, 10(5), 182–186

  • 2021

    Gurram, B. Analysis of Big Data Challenges and Different Analytical Methods. With Reddy, M.D.

    IJERAT, 7(3), 33–38

  • 2021

    Gurram, B. Challenges and Solutions for Improving SSD Performance. Et al.

    IJRES, 9(8), 37–39

03Experience

From medical diagnostics to production clinical AI.

  1. Zasti Inc — AI Engineer

    May 2024 — Present

    Ashburn, VA · Healthcare / clinical AI

    • Architected a production RAPTOR RAG system with hierarchical summarization, Neo4j knowledge-graph integration, and dual-stage vector retrieval (FAISS + Pinecone): 45% faster queries, 38% better factual accuracy, 50M+ document embeddings at sub-100ms P99 latency.
    • Designed a multi-agent orchestration platform (LangGraph, LlamaIndex) with autonomous task planning, tool use, self-reflection loops, and short/long-term memory — automating 95% of clinical research workflows; manual review cycles cut from 3 days to under 4 hours.
    • Engineered a multimodal RAG pipeline (GPT-4V vision, custom OCR post-processors, FHIR-compliant parsers) with hallucination detection via chain-of-thought consistency scoring: 92% retrieval precision across 15+ heterogeneous lab-report formats.
    • Cut LLM inference costs 75% (~$2K/month) via GPTQ 4-bit quantization, speculative decoding, and dynamic batching with vLLM — 98% performance parity on clinical benchmarks (MIRAGE, MedQA).
    • Established AI governance and safety guardrails (RAGAS, LangSmith): 62% lower production hallucination rate, HIPAA-aligned audit logging for all clinical AI outputs.

  2. University of Cincinnati — Graduate Teaching Assistant

    Aug 2022 — Apr 2024

    Cincinnati, OH · Higher education

    • Designed and delivered hands-on workshops in Python, cloud computing, and ML for 500+ graduate students; 90% assignment completion, 95% positive feedback.
    • Mentored 20+ graduate students on capstone research (NLP, CV, predictive analytics); 4 projects selected for departmental showcase, 2 led to published conference abstracts.
    • Built scalable educational resources with responsible-AI principles alongside faculty; +40% engagement, −25% time-to-completion on advanced ML assignments.

  3. Technocolabs Softwares — Deep Learning Developer

    Feb 2021 — Jul 2022

    Remote · Medical diagnostics

    • Led a 4-member team building deep-learning models for a medical-diagnostics product (PyTorch CNNs, ResNet/EfficientNet transfer learning): 88% diagnostic accuracy on held-out clinical test sets, 45% higher system efficiency.
    • Designed a pattern-recognition algorithm combining attention-based feature extraction with ensemble classification: +25% predictive accuracy over baseline for clinical decision support.
    • Built Power BI dashboards with real-time model tracking and SHAP explainability, contributing to a 10% diagnostic-accuracy improvement.

Education

  • M.S. Computer Science · University of Cincinnati

    2022 — 2024

    GPA 3.95/4.0 · Graduate Incentive Award for Academic Excellence ($60,000 scholarship)

  • MicroMasters, Data Science · University of California San Diego

    2021 — 2022

  • B.Tech Computer Science · SRM University (India)

    2018 — 2022

    GPA 3.98/4.0

Toolbox

LLMs & GenAI
LangChain · LangGraph · LlamaIndex · RAPTOR · Self-RAG · QLoRA · DPO · RLHF · vLLM · DSPy
Agentic systems
Multi-agent systems · AutoGen · CrewAI · Tool use · ReAct · Reflection agents · Memory systems
ML / DL
PyTorch · TensorFlow · HuggingFace Transformers · DeepSpeed · FSDP · Scikit-learn · XGBoost
Biomedical
MNE · Nilearn · EEG/fMRI preprocessing · FHIR · Clinical NLP
Vector & graph DBs
FAISS · Pinecone · Qdrant · Weaviate · Milvus · Neo4j · Elasticsearch
MLOps & cloud
AWS SageMaker/EKS/Bedrock · GCP Vertex AI/GKE · Docker · Kubernetes · MLflow · Airflow · FastAPI

04Teaching & mentoring

Research is a team sport — I build the bench.

500+
graduate students taught in hands-on Python, cloud, and ML workshops
20+
capstone research projects mentored across NLP, CV, and predictive analytics
95%
positive student feedback, with 90% assignment completion
4
mentored projects selected for departmental showcase; 2 led to published conference abstracts

In industry

  • Designed 3 technical workshops and a structured internship curriculum for 5 engineers at Zasti (agentic AI, RAG, LLM evaluation).
  • Built adaptive AI learning agents personalizing curriculum content — 40% faster onboarding.
  • Co-developed scalable educational resources with responsible-AI principles alongside University of Cincinnati faculty: +40% engagement, −25% time-to-completion on advanced ML assignments.