Skip to main content
Anuraag Mishra LogoAnuraag Mishra
PRODUCTION AI · RETRIEVAL · RELIABLE SYSTEMS

Anuraag Mishra

Senior AI/ML Engineer

I build production AI systems across retrieval, routing, grounding, semantic reuse, evaluation, and the backend infrastructure that makes them reliable at scale.

Production GenAI & RAGRetrieval & RankingAgentic SystemsAI Reliability & Evaluation
Explore Selected Work ↓Download CV
GitHubLinkedIn|FAU Erlangen-Nürnberg · M.Sc. Data Science
SYSTEM ARCHITECTURE
PRODUCTION AI STACK
FALLBACKQUERY / FILESqueries · documentsUNDERSTAND & ROUTEintent · entities · time · confidenceRETRIEVE & REUSERULESBM25VECTORSCACHEGROUND & REASONscoped context · LLM · toolsEVALUATE & FALL BACKgrounding · freshness · evaluation
Hybrid RetrievalLexical + Vector
Grounded ContextSource-Aware Retrieval
Continuous EvalFallbacks + Quality Checks
CAREER & EDUCATION

Engineering & Research Journey

Software engineering → graduate AI & optimization research → production GenAI systems.

6+ YEARS ENGINEERING
2017·Panjab University

B.E. Computer Science & Engineering

2017–2022·Winshuttle (now Precisely)

Software Engineer

2025·FAU Erlangen-Nürnberg, Germany

M.Sc. Data Science

2025–Present·Synechron · Goldman Sachs engagement

Senior Associate — Technology (AI/ML)

FLAGSHIP SYSTEMS

Featured AI Systems

Three production AI systems showing how I approach routing, grounded document intelligence, and safe semantic reuse.

Explore earlier projects →
Retrieval & RankingEnterprise AI Systems

Adaptive Retrieval & Specialist Routing

Confidence-gated routing for overlapping retrieval paths with a safe default

Enterprise queries can map to overlapping specialist paths, while traffic imbalance makes aggregate accuracy misleading.

  • ›Treat specialist routes as high-confidence overrides instead of forcing every query into a narrow class.
  • ›Combine deterministic, lexical, and semantic evidence, then fall back safely when the routing signal is weak.
Query RoutingBM25Semantic RetrievalEvaluation
Production AICase Study →
Document Intelligence & RAGEnterprise AI Systems

Grounded Document Intelligence

Progressive document ingestion and source-grounded AI over user-controlled collections

Private document collections become harder to search and reason over as ingestion, structure, scale, and source boundaries grow more complex.

  • ›Make documents useful progressively instead of blocking on the slowest ingestion stage.
  • ›Preserve structure, metadata, and source scope so retrieval remains both useful and traceable.
RAGDocument IntelligenceSemantic RetrievalMetadata
Production AICase Study →
AI Reliability & RetrievalEnterprise AI Systems

Semantic Cache & Reliability Layer

Precision-first reuse of previous answers and retrieval results

Repeated and semantically similar requests create unnecessary retrieval and model work, but a false reuse decision can return convincing context for the wrong entity, time, or scope.

  • ›Treat semantic reuse as a retrieval-safety problem rather than nearest-neighbor caching.
  • ›Use entity, time, scope, and freshness constraints to decide whether a candidate can be reused safely.
Semantic CachingHybrid RetrievalQuery UnderstandingNLI
Production AICase Study →
PEER-REVIEWED RESEARCH

Featured Applied Research

Published work combining a compact ReLU network with mixed-integer optimization for tomographic reconstruction. The network learns local edge intensity, and its learned behavior is encoded as optimization constraints to sharpen interfaces while retaining a physically grounded reconstruction objective.

Explore academic work →
JOURNAL ARTICLE · IOP PUBLISHINGNeural Networks × Mathematical Optimization2025
Machine Learning: Science and Technology · 2025

High-quality tomographic image reconstruction integrating neural networks and mathematical optimization

A learned local edge prior encoded as mixed-integer constraints for sparse-view tomography

A compact ReLU network is trained to approximate local edge intensity from image patches, then translated into a mixed-integer formulation and integrated with an optimization-based reconstruction pipeline. The resulting method improves edge sharpness and material homogeneity while preserving explicit control over reconstruction fidelity.

Author Contribution Ownership (Official CRediT Statement)

Lead contribution:Software · Investigation · Formal Analysis · Data Curation
Equal contribution:Conceptualization · Methodology · Validation · Visualization · Writing
PUBLISHED BENCHMARK5 Sparse Projections
CSHM BaselineCSHM 5-projection reconstruction
RME: 0.042
BMS: 0.974
DNN-MIP RODNN-MIP RO 5-projection reconstruction
RME: 0.017
BMS: 0.998

Adapted from Mishra et al., Mach. Learn.: Sci. Technol. (2025) · CC BY 4.0

Mathematical OptimizationNeural NetworksTomographic Reconstruction
View Research PageRead Official Article
EXPERIENCE

Engineering Experience

Building production AI and distributed backend systems across regulated financial workflows and enterprise software.

Senior Associate — Technology (AI/ML)

Synechron · Goldman Sachs engagement · Bengaluru, India

Oct 2025 – Present

Building retrieval, grounding, query-intelligence, and reliability layers for GenAI workflows in a regulated financial environment.

  • ›Designed hybrid retrieval and routing flows that combine deterministic patterns, lexical search, and semantic retrieval so different query types can take the path best suited to them.
  • ›Built query-understanding and reuse layers using semantic caching, entity extraction, and temporal resolution to reduce redundant model work and improve retrieval consistency.
  • ›Developed evaluation and fallback mechanisms for retrieval quality, grounding, freshness, and failure handling so production workflows degrade predictably rather than silently.
PythonFastAPIRAG / LLMsVector SearchBM25Evaluation

Software Engineer

Winshuttle (now Precisely) · India

Jul 2017 – May 2022

Built backend services and distributed data-processing systems for enterprise SAP automation products.

  • ›Developed high-throughput ETL and backend services for processing large volumes of SAP business data across distributed enterprise workflows.
  • ›Built fault-tolerant asynchronous processing with RabbitMQ, retries, dead-letter handling, and restartable workflows for long-running operations.
  • ›Improved processing and delivery reliability through distributed-service design, CI/CD automation, containerization, and production-focused testing.
C# / .NETJavaNode.jsRabbitMQSQLDistributed SystemsCI/CD
LEADERSHIP & PEER ENDORSEMENTS

Professional Recommendations

Direct endorsements from engineering leadership and cross-functional leads on LinkedIn.

View all on LinkedIn
PA

Pallavi Aggarwal

Director of Software Development

Managed Anuraag directly

December 11, 2024
Microservices ArchitectureDistributed SystemsL3 Customer EngineeringTesting & Quality
"Anuraag excelled in taking on complex development tasks, delivering them with impeccable quality, comprehensive unit tests, and thorough integration testing. His versatility was evident in his ability to handle a diverse range of challenges, from working on legacy VB components to RabbitMQ-based solutions. A standout moment was his involvement in rewriting a legacy license management system into a microservices architecture."
Verified on LinkedInRead recommendation
SM

Shelby Murrell

SAP Functional Lead (Winshuttle/Automate Evolve)

Worked with Anuraag on different teams

January 10, 2025
Technical ResolutionCustomer CommunicationCross-Team CollaborationCustomer Engineering
"He was not only gifted in technical skills but also in interpersonal skills. He assisted in making what can typically be a frustrating argument into a clear conversation involving a straightforward resolution. He truly reflects the phrase, 'there is no I in team,' and was always willing to lend his expertise whenever I needed it."
Verified on LinkedInRead recommendation
CORE CAPABILITIES

Technical Capabilities

A working stack across production GenAI, retrieval and ranking, applied machine learning, and distributed backend systems.

GenAI & Agentic Systems

Production LLM systems with grounding, orchestration, evaluation, and tool-aware workflows.

RAGLLMsAgentic AIPromptingGroundingEvaluation

Retrieval & Ranking

Lexical, semantic, and metadata-aware retrieval with query understanding and evaluation.

EmbeddingsBM25Hybrid RetrievalOpenSearchSemantic Caching

ML & Applied Research

Machine learning, mathematical optimization, and scientific experimentation.

PyTorchMathematical OptimizationGurobiComputer VisionExperimentation

Backend & Production Systems

Production APIs, distributed services, asynchronous pipelines, and deployment infrastructure.

PythonFastAPIREST APIsDistributed SystemsRabbitMQDocker / Kubernetes
ARCHIVE

Earlier Projects & Academic Work

Selected earlier projects, graduate seminars, and preserved technical work.

Earlier Projects →FAU Seminar Archive →
CONTACT

Let’s talk about production AI systems.

Open to senior AI/ML engineering roles, applied AI work, and technical conversations around retrieval, agents, evaluation, and reliable AI systems.

anuraagmishra.214@gmail.comConnect on LinkedIn ↗GitHub ↗

Based in Bengaluru, India.