From prototypes to production.
In 2024, I focused on building and deploying enterprise-grade AI solutions across multiple industries, turning prototypes into real systems that deliver measurable impact.
Selected systems and products.
Enterprise AI architectures built for Fortune 500 scale, zero-hallucination tolerance, and deterministic system workflows.
Multi-Agent Procurement Automation Platform
Architected and deployed a multi-agent procurement platform automating vendor selection, purchase-request validation, quotation processing, price extraction, ERP purchase-order creation, and multi-tier approval workflows in an on-premises, air-gapped R&D environment.
- Central purchasing orchestrator coordinating specialist validation/review agents with deterministic price-validation controls.
- Fine-tuned Qwen open-source LLM using LoRA for domain-specific procurement terminology and chemical taxonomy.
- Hybrid enterprise retrieval leveraging Elasticsearch and Milvus for 26 target R&D chemical part numbers.
- Seamless integration with Microsoft Dynamics AX using RPA bridging where direct REST integrations were insufficient.
- Strict air-gapped security compliance covering data sovereignty, IAM/LDAP access, and zero external egress.
Order-to-Cash (OTC) Global Multi-Agent Automation
Engineered an enterprise multi-agent watsonx Orchestrate solution with a global Supervisor Agent intelligently delegating and routing complex business operations across 3 regional specialist agents.
High-Precision Advanced RAG & Discovery Platform
Pioneered a multi-query RAG architecture utilizing 4 parallel discovery strategies (raw query, rephrased query, intent-derived query, and HyDE) integrated with candidate deduplication and Reciprocal Rank Fusion.
IT Incident Intelligence & Dual-Source RAG Assistant
Built an enterprise IT Incident Intelligence Assistant integrating troubleshooting documentation with historical ServiceNow and JIRA incidents via dual-source vector retrieval.
Match Rate & Transformation Analysis Multi-Agent
Cloud-native analytics system on IBM Cloud utilizing a dual-model LLM strategy: Llama 3.2 90B Vision Instruct for multimodal orchestration and Llama 4 Maverick 17B for deep analytical inference.
7-Agent Student Update Process Automation
Engineered an automated 7-agent watsonx Orchestrate + RPA pipeline replacing manual spreadsheet workflows with autonomous ingestion, validation, duplicate detection, and SharePoint updates.
Conversational AI Luxury Shopping Assistant
Developed a conversational shopping agent covering the end-to-end customer journey from semantic catalog search and personalized recommendations to coupon application, payment, and human expert escalation.
Architectural patterns & visuals.
Thin-line system topologies demonstrating real-world multi-agent coordination, hybrid neural retrieval, and air-gapped compliance.
Intent Routing & Asynchronous State
Decomposes high-level transactional requests, validates schemas, and dispatches parallel sub-tasks to specialized domain agents.
How I think & build.
Core engineering philosophy and operational principles that govern how systems are designed, evaluated, and maintained.
“I care less about whether a model can generate a response, and more about whether the system behaves reliably in the real world.”
Generative AI without deterministic boundaries is a prototype. Production engineering is the discipline of wrapping probabilistic models in rigorous type safety, multi-source grounding, strict latency budgets, and transparent governance.
Determinism over stochastic hope
Language models are non-deterministic reasoning engines, not reliable execution engines. Core business logic, database writes, and state mutations must be handled by strongly-typed deterministic code.
Designing reliable multi-agent systems for enterprise Order-to-Cash
Why supervisor-worker agent topologies succeed where autonomous flat swarms fail in complex ERP workflows like SAP and Microsoft Dynamics.
What breaks when naive RAG meets messy unstructured enterprise data
Overcoming document table parsing, OCR degradation, and cross-encoder bottlenecks using Reciprocal Rank Fusion and HyDE query projections.
What AI governance (watsonx.gov) looks like in production environments
Implementing deterministic audit trails, PII scrubbing filters, and human-in-the-loop threshold validation for regulated financial and government systems.
Capabilities organized by discipline.
A structured breakdown of core engineering capabilities without buzzword clouds or arbitrary proficiency percentages.
Agentic AI & Orchestration
Generative AI & LLM Adaptation
Advanced RAG & Vector Intelligence
Enterprise Integration & Cloud Platforms
Recruiter dossier & credentials.
Essential facts, verified metrics, and direct contacts for hiring managers and talent partners evaluating Senior AI Engineer and Architect roles.
Abhishek Singh
AI Engineer and Solutions Architect with 9+ years of software, ML, and AI engineering experience, including 3+ years delivering production enterprise AI systems. Specialized in architecting multi-agent AI platforms, advanced RAG pipelines, open-source LLM adaptation (LoRA), enterprise workflow automation, and air-gapped/cloud deployments for Fortune 500 enterprises including LG Display, Bunge, TfNSW, HDFC, BHP, and ONGC.