[ PRINCIPAL_GENAI_ARCHITECT ]
Open to Senior AI/ML Roles
Hargurjeet Singh Ganger.
Enterprise AI at Scale.
I bridge the gap between proof-of-concept AI models and resilient, production-grade Generative AI architectures. Specializing in enterprise RAG systems and autonomous agentic workflows.
LLMsRAGLangGraphCrewAIAWS BedrockMLOpsPythonOpenSearchFine-tuningVector DBLangChainRagasAgentsQLoRAFastAPIXGBoostGuardrailsLLMsRAGLangGraphCrewAIAWS BedrockMLOpsPythonOpenSearchFine-tuningVector DBLangChainRagasAgentsQLoRAFastAPIXGBoostGuardrails
10+Yrs DS Experience
100K+Files Processed
$50M+Value Realised
memory
70%Reduction in Model Latency
Achieved via custom quantization & RAG pruning.
terminal TECHNICAL_STACK
Generative AI
RAG PatternsAgentsFine-tuningGuardrailsPII FilteringVector DBObservabilityMCP
Frameworks & APIs
LangChainLangGraphCrewAIOpenAI APIAnthropic APIGemini APILangfuseRagas
MLOps & Infra
DockerFastAPIMLflowCI/CDGitHub ActionsGitLabAWS SageMakerFeature Store
Cloud (AWS)
BedrockTextractOpenSearchLambdaStep FunctionsCloudWatchSageMaker
Core ML & Data
XGBoostRandom ForestsPyTorchTensorFlowscikit-learnBERTPySparkSQL
Agentic Tools
CursorClaude CodeKiroAmazon QVS CodeGitHub Copilot
Specializations
Scalable RAG Pipelines, Agentic Workflows, LLM Eval Frameworks, AWS Bedrock.
Philosophy
I bridge proof-of-concept AI with resilient, production-grade architectures.
Featured_Projects
03 PRIMARY / 06 TOTAL2024
AI Agents
Antigravity: Autonomous UI Designer
Autonomous redraw cycles & context retention across agentic iterations using Gemini 2.5.
- Closed-Loop Critic: Triggers autonomous redraw cycles by validating mockup images against guidelines.
- Evaluator Scoring: Computes quantitative metrics for brand consistency, color alignment, and layout accuracy to score attempts.
- State & Memory: Manages agent context and state retention across iterations using a central memory store to refine subsequent generation.
- Robust Guardrails: Enforces strict boundaries via JSON schema validation, retry loops, and degrade-gracefully fallbacks.
- Streaming Timeline: Traces and displays the agentic step-by-step cognitive thoughts and evaluations alongside intermediate drawing cycles.
- Google Agentic Systems: Orchestrates multimodal Gemini 2.5 and Imagen 4 Ultra models to analyze briefs and generate images.
Outcome: A state-of-the-art design workspace demonstrating expert command over multimodal image data through self-correcting agentic loops.
Gemini 2.5Imagen 4Google Agentic SystemsCritique LoopMemoryNext.jsTailwind
2024
AI Agents
Local Multi-Agent Folder Organizer
A local-first system running fully on-device via Ollama that restructures cluttered downloads folders into semantic, context-aware nested subdirectories.
- Hierarchical Architecture: Configures a lead orchestrator agent that partitions folder listings into subtask categories, preventing context window limits.
- Multi-Agent Concurrency: Spawns category-specific specialist subagents in parallel using a Python ThreadPoolExecutor to slash local Ollama inference latency by 60%.
- Pydantic Output Validation: Enforces strict JSON schemas on local SLMs using CrewAI's output parsing, guaranteeing zero formatting errors.
- Human-in-the-Loop Safe Gate: Implements a CLI preview table and user confirmation prompt before mutating any folder structure, supporting a dry-run mode.
- Transactional Rollback Log: Records all file migrations atomically in a central history.json transaction log, facilitating instant programmatic recovery.
Outcome: A local-first system running fully on-device via Ollama that restructures cluttered downloads folders into semantic, context-aware nested subdirectories.
CrewAIOllamaLlama 3.2PydanticPythonLocal LLMsSystems Automation
2024
AI & Benchmarks
Local AI Assistant & SLM Benchmarking
Rigorous local benchmark of 30 multi-domain prompts published on Dev.to and GitHub. Proved Llama 3.2 (3B) is the most reliable for structured JSON.
- Local SLM evaluation: Developed a FastAPI testing harness benchmarking Llama 3.2 (3B), Phi-3 Mini (3.8B), and Mistral (7B) fully on-device via Ollama.
- Inference speed profiling: Measured performance where Phi-3 Mini led at 22.70 tokens/sec (323.99ms TTFT), and Llama 3.2 followed at 22.24 tokens/sec (427.29ms TTFT).
- Pydantic schema enforcement: Structured LLM outputs using validation schemas. Llama 3.2 achieved 100% compliance via retry reprompts, and Mistral 7B achieved 90% compliance.
- Resource allocation tracking: Measured memory-bound constraints on Apple Silicon Mac mini (16GB RAM) where CPU load remained low (13-15%) but memory hit 88.8% to 94.4%.
Outcome: Rigorous local benchmark of 30 multi-domain prompts. Proved Llama 3.2 (3B) is the most reliable for structured JSON pipelines, while Phi-3 Mini excels in speed.
OllamaFastAPILlama 3.2Mistral 7BPhi-3PydanticPythonApple Silicon
Experience_Log
TRANSITION FROM QUALITY ASSURANCE
TO PRODUCTION AI SYSTEMS BUILDER.
TO PRODUCTION AI SYSTEMS BUILDER.
May 2022 – Present
- Conversational AI & RAG Architecture: Architected and led delivery of an enterprise-grade conversational AI system (LLMs + RAG), enabling natural language queries at scale and driving a 70% reduction in manual data extraction time, while mentoring a cross-functional team across design, deployment, and production rollout. Built a multimodal document pipeline using AWS Textract, pdfplumber, and OpenSearch to process 100K+ documents (PDFs, images, scanned files) at 90%+ accuracy, integrating AWS Bedrock for real-time LLM inference with p50/p95/p99 latency tracking.
- Agentic Workflows & Production Guardrails: Designed and implemented multi-step agentic workflows using CrewAI and LangGraph with tool-augmented pipelines, JSON schema validation, retry loops on malformed LLM output, and graceful degradation — enforcing hallucination guardrails at production scale. Deployed on AWS with Docker containerisation and GitLab CI/CD, with full Langfuse observability for end-to-end LLM tracing and token usage monitoring.
- Email Intelligence Pipeline & Fine-Tuning: Designed an automated email intelligence pipeline processing 6,000+ weekly escalation emails, eliminating manual triage and enabling real-time ticket creation in ServiceNow via AWS Lambda, Step Functions, and Microsoft Graph API. Fine-tuned LLaMA-2 7B locally using QLoRA (r=16, 4-bit NF4 quantisation) on 3,000+ annotated emails, achieving a 40% F1-score improvement over baseline prompting — deploying custom LoRA adapters to AWS Bedrock for production inference.
- Multi-Agent Deep Research System: Architected a multi-agent deep search system featuring a Planning Agent that decomposes complex research queries into structured sub-tasks, parallel Search Agents that execute targeted retrieval across sources, and a Synthesis Agent that aggregates, deduplicates, and distills findings into coherent final reports — enabling autonomous, multi-hop research at scale.
- Recommendation Systems & Revenue Growth: Engineered a suite of recommendation systems to drive revenue across BT's product portfolio — built a multi-label classifier (Random Forest + XGBoost) to identify and upsell premium SD-WAN products, increasing sales by 10%, and implemented a market basket analysis pipeline (Apriori) to uncover cross-sell patterns for Value-Added Services (VAS), achieving a 30% uplift in VAS sales.
Academic_Log
FORMAL EDUCATION & SPECIALIZED CERTIFICATIONS.