PB / AI Engineer
Experience Projects Skills Writing Contact
Artificial Intelligence Engineer · Gen AI Systems
PRIYANSH
BHARDWAJ

I build production AI systems - RAG pipelines, recommendation engines, and GenAI infrastructure - that go beyond proof-of-concept into reliable, measurable, scalable software. across fintech, retail analytics, and healthcare, working at the boundary between ML research and engineering discipline.

Current Role
Senior Applied Data Scientist
Dunnhumby · Apr 2025 – Present ·
Total Experience

AI Engineering · GenAI · MLOps
Domain Expertise
Gen AI Systems · RAG · LLMOps
Recommendation · Summarisation
Work Eligibility
Open to Canada, EU & UK roles
Visa sponsorship required · Relocation ready
View Projects Experience GitHub LinkedIn
Scroll
01 - Experience

Career Timeline

Three roles across fintech, retail analytics, and independent product engineering - each ending with shipped, monitored systems in production.
Apr 2025 - Present · Current
Dunnhumby
Senior Applied Data Scientist

Dunnhumby is a global customer data science company serving major retailers including Tesco, Kroger, and Meijer. My remit covers the full pipeline from raw data and LLM architecture to production deployment and ongoing evaluation. Enterprise-scale work: large internal knowledge bases, multiple teams consuming AI outputs, and reliability expectations that make experimentation-mode thinking inadequate.

  • Architected an enterprise RAG system for natural language access to internal knowledge bases - 92% answer accuracy, hallucination rate below 5%, 60% reduction in manual search effort.
  • Designed the microservices AI architecture - distributed FastAPI services, Docker + Kubernetes, horizontal autoscaling. Retrieval and generation optimisation reduced end-to-end latency by 10%.
  • Built evaluation infrastructure measuring context relevance, faithfulness, and response quality continuously - automated metrics with alerting on degradation.
  • Productionised AI summarisation and structured query agents - reducing manual processing workload by 65%.
  • Led cross-functional collaboration between data engineering and product teams to transition prototypes into stable, documented production systems.
RAGFastAPIKubernetesDockerLLMsRAGASEvaluation PipelinesAI AgentsMicroservices
Sep 2022 - Mar 2025 · 2 yrs 7 mosIndependent
Independent Product Development
AI / ML Engineer

Built and shipped three end-to-end AI products independently across healthcare, legal, and fintech - each with a deliberate evaluation methodology before claiming results. Self-directed accountability to hard metrics without the safety net of a team to validate assumptions.

  • Built a production RAG system for healthcare retrieval - 95% precision via RAGAS evaluation, hallucination monitoring, BioBERT embeddings.
  • Developed an embedding-based recommendation engine using two-tower architecture - ROC-AUC from 0.82 to 0.90+ through better embeddings and hard negative sampling.
  • Delivered a legal summarisation platform with 92% clause preservation - map-reduce LLM summarisation with cosine similarity evaluation.
  • Built full MLOps stack: CI/CD via GitHub Actions, drift detection, Prometheus and Grafana dashboards.
FAISSPineconeWeaviateLangChainLlamaIndexBioBERTMLflowPrometheus
Sep 2021 - Jun 2022 · 10 mosFirst Role
Signzy
AI Engineer

Signzy provides AI-powered KYC and financial onboarding infrastructure to banks and fintechs. First engineering role on live transaction pipelines where production errors have direct financial consequences.

  • Built fraud detection models on live financial pipelines - 40% reduction in manual review, 40% cut in false positives.
  • Automated model training and deployment workflows - 80% reduction in operational overhead, 30% in processing time.
  • Contributed to scalable ML infrastructure for automated orchestration and production monitoring.
Fraud DetectionML PipelinesAutomationFintechKYC
02 - Projects

Selected Work

Five production systems across healthcare, fintech, and legal. Every cited metric comes from an evaluation pipeline - not a back-of-envelope estimate.
01 - Featured · Healthcare Domain

Healthcare RAG
Retrieval System

Natural language querying over thousands of medical documents, returning cited, verifiable answers grounded in the source corpus. Core challenge: eliminating hallucination without sacrificing coverage or latency.

Achieved 95% summarisation precision and hallucination rate below 5% - RAGAS evaluation on a stratified 200-query test set supplemented by human annotation.
Stack
PythonLangChainWeaviateFAISSBioBERTFastAPIRAGASDockerPrometheus
Architecture Decisions That Mattered
Parent-child chunking
256-token child for precision, 1024-token parent to LLM. Context Precision: 0.71 to 0.87.
BioBERT domain embeddings
Myocardial infarction and heart attack are near-identical vectors. General models treat them as distinct - dangerous in clinical retrieval.
Two-stage re-ranking
Bi-encoder retrieves top-50, cross-encoder re-ranks to top-5. ~100ms cost eliminates irrelevant context that causes hallucination.
NLI faithfulness scoring
Every response claim scored against retrieved chunks via NLI before delivery. Per-query monitoring, not sampling.
Hybrid BM25 + dense search
Drug names and ICD codes fail on semantic search alone. Both modes required for clinical language coverage.
02

Embedding-Based
Recommendation Engine

Two-tower dense retrieval for fintech - users and items as dense vectors via ANN. Solved cold-start where collaborative filtering failed. Redis-cached embeddings, sub-50ms P99 latency.

ROC-AUC 0.82 to 0.90+ · Sub-50ms P99
FAISSPineconeSentence TransformersRedisFastAPI
03

Legal Document
Summarisation Platform

Map-reduce LLM summarisation with clause-level semantic evaluation. A missed clause is a liability. Preservation measured via cosine similarity, calibrated on 50 human-labelled examples.

92% clause preservation · Containerised
LangChainSentence TransformersFastAPIDocker
04–05

Time-Series
Forecasting Systems

Stock price forecasting with Kafka streaming and Kubernetes autoscaling; AQI prediction covering 300+ Indian locations with real-time dashboards. Both cross-validated and deployed.

92% CV accuracy · 94% accuracy (300+ locations)
XGBoostKafkaKubernetesStreamlit
03 - Skills

Technical Depth

Visualised by actual usage depth in production systems - not self-assessed familiarity from tutorials.
Languages
PythonExpert
SQLAdvanced
Bash / ShellProficient
AI / ML
RAG Systems & LLMsExpert
Vector Search - FAISS, Pinecone, WeaviateExpert
NLP & Text EmbeddingsExpert
ML Evaluation & RAGASExpert
Classical ML - XGBoost, SklearnAdvanced
Recommendation SystemsAdvanced
Deep Learning - PyTorch, TensorFlowAdvanced
MLOps & Infrastructure
FastAPI & REST API DesignExpert
Docker & KubernetesAdvanced
CI/CD, GitHub ActionsAdvanced
Prometheus, Grafana, MLflowProficient
Cloud - AWS, GCP, AzureProficient
GenAI Frameworks
LangChainLlamaIndexLangGraphHuggingFaceRAGASSentence TransformersOllamaOpenAI APIAnthropic APIChroma
Data Engineering
KafkaAirflowSnowflakePySparkdbtPostgreSQLMongoDBRedisPandasNumPy
Experience by Domain
Certifications
Weaviate
Vector DB Professional
DeepLearning.AI / Coursera
Deep Learning Specialisation
DeepLearning.AI / Coursera
Gen AI for Software Development
DeepLearning.AI / Coursera
CNNs in TensorFlow
DeepLearning.AI / Coursera
Intro to TensorFlow for AI, ML & DL
LinkedIn Profile
View all certificates →
Education
Degree

B.Tech, Computer Science & Engineering
Modern Institute of Technology & Research Centre, Alwar
2018 – 2022 · CGPA 8.95 / 10

Academic Standing

CGPA 8.95 within the top percentile of the graduating cohort. Strong foundation in algorithms, data structures, applied mathematics, and distributed systems.

International Recognition

Degree recognised at H+ level on the German Anabin database. Qualifies for EU Blue Card in Germany, Global Talent Stream(GTS) and Express Entry in Canada and Skilled Worker Visa across the EU and UK. Open to relocation. Visa sponsorship required.

04 - Writing

Blog & Publications

Technical writing on applied ML - production trade-offs, evaluation methodology, and honest accounts of what the numbers mean.
Published · Medium · Jul 2024

Predicting Air Quality with Machine Learning

A complete walkthrough of building a 94%-accurate AQI system for 300+ Indian locations - feature engineering, model selection, and what working with real government monitoring data actually looks like.

Read on Medium →
Published · TDS · April 2026

Grounding Your LLM: A Practical Guide to RAG for Enterprise Knowledge Bases

What a successful RAG demo hides - and what breaks the moment someone asks about real company data. A production walkthrough from architecture to the evaluation step most teams skip entirely.

Read on Towards Data Science (TDS) →
Published · TDS · April 2026

Your Chunks Failed Your RAG in Production

A post-mortem on a specific chunking failure in production - what went wrong, why it passed testing, and what the fix actually looked like. The kind of thing you only learn by shipping, not by reading docs.

Read on Towards Data Science (TDS) →
05 - Contact

Let's Connect

Open to senior AI/GenAI Engineering roles in Canada, EU and UK. Interested in teams where evaluation rigour and production reliability are first-class concerns.

Available for new opportunities

Actively looking for senior AI or Gen AI Engineering roles in Germany, Canada, the Netherlands, the UK, and the EU broadly. Happy to talk through relocation timelines, visa support, and what the day-to-day looks like before either of us commits.