App LogoApp name
AP

Anuj Pratap Singh

Open to work3 years experience

AI / ML Engineer

Noida, IndiaTarget Roles: Backend Engineer • Full-Stack Developer • Frontend Specialist

AI / ML Engineer – Applied GenAI & Agentic Systems

GitHub

Standing Rank

Rank Not Available

Developer Badges

No badges earned yet

Skills & Technologies

47 skills
langchainlanggraphtemporal.ioragprompt engineeringagentic workflowsguardrails aistructured output designopenai apiazure openaianthropic apitransformershuggingfacedonutlayoutlmv3whisperfine-tuningchromadbqdrantelasticsearchbge-m3bge-rerankerhybrid searchazure document intelligenceocrcamelotpymupdflabel studiotesseractopencvarize phoenixragaspydanticlangsmithpythonfastapijavaspring bootreact.jsreact nativenode.jsjavascriptdockergitmysqlpostgresqlrest apis

Work Experience

Senior Software Engineer (AI/ML & GenAI Specialist)

Newgen Software

Jan 2023 - Present Noida, India
  • Architected and deployed an agentic Retail Loan Origination System (RLOS) using LangGraph + Temporal.io, orchestrating durable, human-in-the-loop workflows across 5 regions.
  • Built and productionized an Intelligent Document Processing (IDP) pipeline for trade finance using Azure Document Intelligence + OpenAI, automating extraction across 25+ document types at scale.
  • Delivered enterprise banking/insurance applications on Java, Spring Boot, React and Newgen low-code platforms (OmniFlow/iBPS, OmniDocs), leading a small team through Agile delivery cycles.

Projects

Trade Finance Intelligent Document Processing – (IDP)

Azure AI FoundryOpenAISpring BootRabbitMQDockerNewgen IBPS
  • Deployed a live production IDP pipeline for a UAE bank, automating classification and field extraction across 25+ document types (invoices, bills of lading, letters of credit) at hundreds of documents/day.
  • Combined Azure OCR with strict JSON-returning OpenAI prompts, zone-based verification, and rule-based post-processing; integrated via RabbitMQ for asynchronous, multi-threaded, high-throughput ingestion.
  • Impact:Cut manual document processing time per case by an estimated 70%, reducing average turnaround from~15 minutes to under 3 minutes per document.

Credit Card Application Extraction – Donut Transformer

HuggingFaceDonutLabel StudioPython
  • Built an OCR-free document understanding POC for RHB Bank, fine-tuning a HuggingFace Donut transformer to extract 35+ fields directly from offline credit card application form images.
  • Achieved 80% field-level accuracy while eliminating external OCR/LLM API dependency entirely – satisfying strict data-privacy requirements via fully on-device inference.
  • Designed evaluation strategy using field-level exact match, token-level F1, and a human-in-the-loop review layer for iterative dataset expansion.
  • Impact:Reduced manual data-entry effort per application by~75%, saving an estimated 8–10 minutes of back-office keying time per form.

Relationship Manager Call Intelligence Agent

LangGraphOpenAI WhisperTwilioGuardrails AIArize PhoenixPydantic
  • Built a dual-mode conversational AI agent transcribing live Twilio calls and uploaded recordings via Whisper, feeding a LangGraph extraction pipeline for intent, borrower details, and cross-sell signals.
  • Combined an ML decision engine with LLM-based lead scoring to classify leads as Hot/Warm/Cold; enforced output compliance with Guardrails AI and full trace observability with Arize Phoenix.
  • Impact:Cut average call-review time from~20 minutes to under 5 minutes per call (~75% faster), speeding up lead follow-up and prioritization.

RFP Response Agent – On-Premises LLM

Mistral-Nemo-7BQdrantElasticsearchBGE-M3/RerankerRAGASPhoenix
  • Built a hybrid retrieval pipeline (Qdrant dense + Elasticsearch keyword search, BGE-Reranker cross-encoding, query rewriting) to auto-draft RFP responses from historical documents.
  • Deployed Mistral-Nemo-7B on-premises for data privacy, sizing compute via VRAM/quantization/KV-cache calculations; reduced drafting time from 1–2 days to under 2 hours (~85% reduction), validated with a RAGAS + human evaluation stack.

Agentic Portal Builder with Embedded AI Layer

ReactSpring BootLangGraphAntigravityPostgreSQL
  • Designed an agentic AI layer on top of a JSON-schema-driven full-stack portal builder, exposing 8 schema-mapped tools through LangGraph with PostgreSQL-backed checkpointing for stateful, resumable agent sessions.
  • Built a dedicated Portal Builder API for direct Spring Boot REST integration, enabling natural-language-driven portal/workflow generation on an existing low-code platform.
  • Impact:Reduced portal/workflow configuration time from days to hours, an estimated~60% faster delivery versus manual low-code setup.

Agentic Cash Finance Journey – Multi-Channel

LangGraphTemporal.ioVoice + Chat
  • Designed a durable, agentic Cash Finance customer journey using LangGraph for reasoning/orchestration and Temporal.io for saga-based, fault-tolerant workflow execution.
  • Supported both chat and voice interaction modes with multilingual system prompts (Arabic/English/Hindi), modeled on a real Saudi Cash Finance workflow.
  • Impact:Automated multi-step verification and approval logic, cutting estimated manual application-handling time by ~50–60%.

CLOS Document Intelligence – Loan/Credit Agreements

RAGChromaDBPydanticOpenAIArize Phoenix
  • Built a RAG-based extraction pipeline over CLOS documents (loan agreements, issuing bank agreements, revolving credit facilities), using ChromaDB for retrieval and Pydantic for strict structured-output validation.
  • Instrumented Arize Phoenix for end-to-end LLM observability and trace-level accountability of extracted fields.
  • Impact:Reduced manual review time per agreement from~30 minutes to under 5 minutes (~80% faster), improving extraction consistency across document types.

Leaderboard Standings

Leaderboard Position Pending

Global test scores, peer standing percentiles, and algorithm leaderboard ranks are updated dynamically.

Assessment Highlights

Assessments Not Completed

Coding evaluations, system assessment results, and conceptual score badges will appear here after taking a test.

AI Collaboration Score

AI Collaboration Score Pending

Developer coding behavior, assistant cooperation, and AI pair-programming indicators are evaluated during live coding sessions.

Role Compatibility Profile

Role Compatibility Analysis Pending

Custom matching reports, candidate role compatibility percentiles, and core engineer strength profiles are processed once conceptual code screenings are complete.

Achievements

Achievements Not Earned

Special rewards, developer badges, system recognition certificates, and conceptual milestones will display here.

About Details

Professional Bio

AI/ML Engineer with 3.6+ years of software engineering experience, specializing in production-grade Generative AI, multi-agent systems, and Advanced RAG architectures. Combines full-stack proficiency with deep LLM infrastructure knowledge to transition complex AI systems to enterprise deployment.

B.Tech

JSS Academy of Technical Education, Noida, Uttar Pradesh ( - 2023)

Languages: English, Hindi
Anuj Pratap Singh - Profile | Swiftcruit