// AI & AGENTIC SYSTEMS ENGINEER

Akshat
Chaudhary

I build autonomous AI systems that actually run in production — vision-based layout detection, voice agents, graph AI, multi-agent orchestrators, RAG pipelines, and the infrastructure that keeps them alive. Moving beyond wrapper agents into deep system integration — Graph DBs, dynamic SQL, computer vision. Not demos. Deployments.

0
PRODUCTION DEPLOYMENTS
0%
PEAK SYSTEM ACCURACY
VOICE AGENT LATENCY
0K+
DOCUMENTS PROCESSED
01 // DEPLOYMENT LOG

Where the agents
run live.

SYSTEM ONLINE — JAN 2026 → PRESENT
AI Engineer
RChilli Inc.  ·  Chandigarh, India  ·  Full-time
01
Designed and implemented a YOLOv11-based PDF Layout Detection system to accurately extract structured and unstructured text from complex documents — multi-column layouts, tables, headers, footnotes
02
Prepared and annotated large-scale training datasets (2.4K+ images) using Roboflow, enabling high-precision layout classification and bounding-box detection
03
Developed an AI-powered Candidate Pre-Screening Agent integrating ElevenLabs API for real-time voice-based interviews and GPT-4 for automated candidate evaluation and decisioning
04
Engineered a Neo4j Graph AI Agent to map complex entity relationships (job-to-job, skill-to-job, skill-to-skill) using custom graph node architecture over the Neo4j platform
05
Architected DML (Central Brain Engine) — a reasoning engine using hybrid vector search and dynamic SQL generation to feed real-time database records directly into AI operations, bypassing wrapper-agent limitations
06
Optimized ML models for CPU inference and dockerized services using FastAPI and Gunicorn for scalable, environment-agnostic deployments across environments
ARCHIVED — APRIL 2024 → DEC 2025
AI & Agentic AI Engineer
TwoPir Consulting  ·  Chandigarh, India  ·  Full-time
01
Delivered end-to-end AI automation using LangChain agents, GPT-4/Claude APIs, and multi-agent orchestration, reducing manual intervention by ~40%
02
Integrated Typeform, Slack, HubSpot via OpenAI APIs into conversational automation pipelines — improved client engagement by ~30%
03
Managed Python (Pandas, NumPy) and SQL/NoSQL data pipelines, improving model input quality by ~25%
04
Cloud & MLOps: deployed inference on AWS Lambda, ECS, S3 using Docker + Terraform for CI/CD automation
05
Collaborated with international stakeholders to architect AI-first automation solutions with measurable ROI improvements across accounts
02 // DEPLOYED SYSTEMS

Nine systems,
running in prod.

STATUS: 8/9 LIVE
UPTIME: 99.9%
LIVE
SVC-001 / JAN–MAR 2025
WebRTC Voice Agent
with GPT-4o Realtime
Enterprise-grade conversational AI with real-time voice, contextual memory, and bi-directional CRM sync. Not a chatbot — an actual phone agent.
95%
ACCURACY
<800ms
LATENCY
+45%
ENGAGEMENT
GPT-4o Realtime WebRTC Twilio ElevenLabs HubSpot AWS Lambda Redis
VIEW CODE
const voiceAgent = async () => {
  const agent = new AdvancedAIAgent({
    model: 'gpt-4o-realtime',
    voice: 'neural-tts-streaming',
    tools: [intelligentCRMSync, contextualMemory],
    memory: new MultiModalConversationMemory(),
  });
  await agent.connect(twilioStream);
  return agent;
};
VIEW CASE STUDY
LIVE
SVC-002 / AUG–NOV 2024
SmartReports — Multilingual
RAG Intelligence Engine
Multilingual reporting system using RAG + semantic search. Evaluates document repositories, extracts intelligence, and auto-generates reports for 6 language markets.
95%
ML ACCURACY
-50%
MANUAL OPS
6
LANGUAGES
Hugging Face Pinecone LangChain OpenAI n8n DeepL AWS S3
VIEW CODE
const ragEngine = new AdvancedRAGPipeline({
  embeddings: huggingFaceEmbeddings,
  vectorStore: pineconeVectorDB,
  llm: openai.gpt4Turbo,
  strategies: ['semantic-search','chunk-reranking']
});
const report = await ragEngine.synthesize({
  languages: ['en','fr','nl','de','es','pt'],
});
VIEW CASE STUDY
LIVE
SVC-003 / JUN–SEP 2024
Multi-Agent CRM
Intelligence Platform
AI-orchestrated HubSpot automation — intelligent agents handle queries, predictive lead scoring, webhook-driven workflows, and autonomous meeting scheduling.
-85%
RESPONSE TIME
+42%
CONVERSION
-70%
MANUAL TASKS
GPT-4 HubSpot API n8n Webhooks LangGraph
VIEW CODE
const agent = new MultiStrategyIntelligentAgent({
  crm: hubspotAPI,
  llm: openai.gpt4Turbo,
  automation: n8nWebhookOrchestrator,
  strategies: ['intent-classification',
               'predictive-scoring'],
});
await agent.processWithIntelligence(query)
           .executeWorkflow();
VIEW CASE STUDY
LIVE
SVC-004 / JAN–MAR 2024
Enterprise RAG
Customer Support Agent
Multi-stage RAG with hierarchical retrieval and context reranking. Handles 100K+ documents, persistent memory across sessions, 92% query resolution rate.
92%
RESOLUTION
-60%
RESPONSE TIME
100K+
DOCS INDEXED
LangChain Pinecone GPT-4 Hugging Face FastAPI Python
VIEW CODE
const rag = new MultiStageRAGAgent({
  vectorDB: pineconeMultiIndex,
  embeddings: huggingFaceSemanticModel,
  memory: persistentConversationalMemory,
  strategies: ['semantic-chunking',
               'hierarchical-retrieval'],
});
await rag.evaluateDocumentsConsecutively(query)
         .synthesizeResponse();
VIEW CASE STUDY
LIVE
SVC-005 / JAN 2026
Vision-Based PDF
Layout Detection Engine
YOLO-powered computer vision pipeline for extracting and semantically classifying text from complex multi-column document layouts and tables. Applies LightSVM on extracted embeddings for semantic section assignment.
2.4K+
TRAIN IMAGES
CPU
OPTIMIZED
100%
DOCKERIZED
YOLOv11 Sentence Transformers LightSVM PyMuPDF pdfplumber FastAPI Roboflow
VIEW CODE
model = YOLOv11(weights='layout-detect-v3.pt')
results = model.predict(pdf_image, conf=0.45)
regions = extract_regions(results, strategy='bbox-merge')
embeddings = sentence_transformer.encode(regions)
labels = lightSVM.classify(embeddings)
return semantic_sections(labels, regions)
VIEW CASE STUDY
LIVE
SVC-006 / FEB 2026
Autonomous Voice
Pre-Screening Agent
Multi-agent LLM architecture conducting real-time, voice-based candidate interviews, analyzing transcripts, and automating HR hiring decisions (Recommended / Not Recommended) via secure webhooks. No human in the loop.
GPT-4
CORE MODEL
Voice
INTERFACE
Auto
DECISIONING
OpenAI SDK GPT-4 ElevenLabs API FastAPI Python Webhooks
VIEW CODE
agent = PreScreeningAgent({
  llm: openai.gpt4,
  voice: elevenLabsAPI,
  decision_engine: HiringDecisionModel(),
  output: webhookDispatcher,
})
result = await agent.conduct_interview(candidate)
               .evaluate_transcript()
               .dispatch_decision()
VIEW CASE STUDY
LIVE
SVC-007 / MAR 2026
DML — Central Brain
Database Engine
Core reasoning engine that fuels AI agents with real-time operational data. Utilizes hybrid vector search and dynamically generates SQL queries to extract exact table records — bypassing superficial wrapper-agent limitations entirely.
Hybrid
VECTOR SEARCH
Live
SQL QUERIES
Real-time
DATA-DRIVEN
Dynamic SQL Hybrid Search Vector DB Agentic AI Python
VIEW CODE
engine = DMLBrainEngine({
  vector_search: hybridSearchLayer,
  sql_generator: DynamicSQLEngine(),
  data_source: operationalDB,
})
records = engine.retrieve(query, mode='hybrid')
result = engine.inject_to_agent(agent, records)
VIEW CASE STUDY
IN-PROG
SVC-008 / EARLY 2026
Neo4j Relational
Graph AI Agent
Graph-based AI agent that maps and traverses complex entity relationships between jobs and skills using real-world data. Custom node architecture over Neo4j enables multi-hop reasoning and skill-gap analysis at scale.
Graph
DATABASE
Entity
MAPPING
Live
TRAVERSAL
Neo4j Cypher Graph AI Python LLM Orchestration
VIEW CODE
graph_agent = Neo4jAIAgent({
  driver: neo4j.driver(NEO4J_URI),
  llm: openai.gpt4,
  node_schema: EntityRelationshipSchema,
})
graph = graph_agent.map_entities(job_data, skill_data)
path = graph_agent.traverse(src='python', dst='ml-engineer')
VIEW CASE STUDY
LIVE
SVC-009 / 2025
Interview Intelligence:
AI Audio Analyst
End-to-end interview analysis system that transcribes audio via OpenAI Whisper, then evaluates candidate responses across 5 behavioral dimensions using LLaMA 3.1 8B on Groq. Delivers structured JSON reports in under 2 minutes.
95%+
ACCURACY
<2min
REPORT TIME
5
SCORE DIMS
OpenAI Whisper LLaMA 3.1 8B Groq API FastAPI Next.js
VIEW CASE STUDY
03 // AGENT STACK

The pipeline that
makes agents real.

// HOW AN AI AGENT ACTUALLY WORKS — AND WHERE I FIT IN EACH LAYER
USER INPUT HUGGING FACE EMBEDDINGS PINECONE VECTOR SEARCH GPT-4o · CLAUDE LLM REASONING LANGGRAPH AGENT n8n · CRM · AWS TOOL CALLS AUTOMATED OUTCOME
LLM LAYER
  • GPT-4o Realtime
  • Claude APIs
  • Llama / Mistral
  • Fine-tuning
  • Prompt Engineering
  • YOLOv10 / YOLOv11
  • Sentence Transformers
  • LightSVM / LogReg
AGENTIC
  • LangChain Agents
  • LangGraph
  • Multi-agent Orch.
  • AutoGen / CrewAI
  • Tool Use & Memory
  • ElevenLabs Voice AI
  • Neo4j Graph DB
  • Roboflow
RAG & DATA
  • Pinecone / Weaviate
  • Hugging Face
  • Semantic Search
  • Pandas / NumPy
  • ETL Pipelines
  • PyMuPDF / pdfplumber
  • Hybrid Vector Search
  • Dynamic SQL Gen
CLOUD & OPS
  • AWS Lambda/ECS/S3
  • Docker / Terraform
  • CI/CD Pipelines
  • CloudWatch / IAM
  • Azure
  • Gunicorn
  • FastAPI
AUTOMATION
  • n8n Workflows
  • Airflow DAGs
  • Webhook Systems
  • CRM Integrations
  • Event-driven Arch.
BACKEND
  • Python / FastAPI
  • TypeScript
  • PostgreSQL / MongoDB
  • REST APIs
  • Salesforce
04 // CREDENTIALS

Certified where
it matters.

Bachelor of Technology — Computer Science
Chandigarh Engineering College, Mohali, India  ·  June 2020 – August 2024  ·  GPA: 8.1 / 10
SF
Salesforce Agentforce AI Specialist
Salesforce  ·  March 2025
★ CERTIFIED — Rare credential, <2025 cohort
SF
Salesforce Certified Platform Developer I
Salesforce  ·  December 2024
⬡ Trailhead Ranger — 100+ Badges, 100K+ Points
AI
AI Foundations: Prompt Engineering with LLMs
Coursera  ·  June 2025
NS
Artificial Intelligence
SkillUp – NASSCOM  ·  August 2024
AWS
AWS Educate Cloud Foundations
Amazon Web Services  ·  July 2024
05 // INITIALIZE CONNECTION

Ship something
autonomous.

Open to senior AI engineering roles, ambitious freelance builds, and anyone who wants to put real agents into production. Mention EASTEREGG in the terminal for something special.

$ ping akshat --channel email
$ ping akshat --channel phone
$ ping akshat --channel linkedin
$ // or open the terminal and type 'hire' 👀
akshat@terminal:~$
Welcome to Akshat's Terminal v2.0.0 Type 'help' for available commands. System online. All agents operational.
$