Kartikey Sharma — AI Agent Developer

Work History

Where I've
Built & Contributed

Engineered LangGraph-orchestrated tutoring agents with RAG over course corpora (semantic chunking + top-k retrieval, k=5), improving grounded answers vs single-shot prompting and reducing off-topic/low-grounding responses by ˜25–40% on internal evaluation-style checks.

Integrated two production LLM backends (GPT-4o, Groq LLaMA) behind JSON Schema–validated outputs, cutting malformed generations by ˜35–55% and reducing retry/cleanup cycles by ˜20–35% for downstream consumers.

Optimized blended cost and latency via tiered routing: steered ˜60–75% of high-volume classroom traffic to Groq while reserving GPT-4o for complex reasoning, typically lowering median latency ˜30–50% and reducing blended LLM spend ˜30–45% vs an always GPT-4o baseline on comparable workloads.

Building and operating live agentic systems for automated patent drafting in production — extracting technical claims and generating structured patent language aligned with IP filing standards, with ongoing iteration from real usage.

Running multi-step LangGraph agents that parse invention disclosures, assess novelty, and produce legally structured drafts — tuned continuously with IP stakeholders for accuracy in high-stakes workflows.

Serving RAG over legal and technical corpora so outputs stay grounded in prior-art context and domain terminology; monitoring quality and adjusting retrieval and prompts as the product evolves.

Integrating LLM backends with structured outputs and validation at the boundary so production runs stay reliable while the drafting pipeline scales with active users.

Shipping AI features on the live Celebso platform — India's premium creator and innovator hub — so creators, founders, and professionals can grow visibility and influence with production-grade reliability.

Running intelligent profile recommendation and discovery (semantic search + embeddings) in production to connect talent with investors and industry leaders as users join and iterate.

Operating AI agents for startup pitch analysis, talent evaluation, and event workflows across the ecosystem, with continuous tuning from real user behavior.

Integrating LLMs to personalize the live product experience — helping users build stronger professional brands and get discovered faster. (celebso.com)

Designed and shipped FastAPI/Flask ML services for classification and recommendation, separating preprocessing → inference → post-validation to stabilize APIs and speed iteration (˜15–25% fewer cross-layer regressions vs mixed monolith-style handlers, typical for layered serving).

Productionized an end-to-end inference pipeline (feature prep, model serving, synchronous response validation) for low-latency model-backed endpoints, targeting ˜200–400ms p95 for lightweight requests (model-dependent).

Implemented schema + rule-based guardrails and automated response checks at the boundary, reducing invalid outputs reaching clients by ˜20–35% and cutting downstream failure incidents tied to bad responses by ˜15–30% (typical when validation is moved earlier).

Added defensive reliability patterns (timeouts, safe defaults, graceful degradation), improving stability during provider variability and bursts—typically ˜20–35% fewer user-visible hard failures vs no-timeout/no-fallback paths in similar systems.

Built ML-assisted anomaly checks on large-scale government datasets to flag inconsistencies earlier in QA, reducing manual review cycles by ˜25–40% for recurring quality gates before reporting.

Automated reporting + validation workflows to standardize checks and strengthen audit readiness, cutting repetitive manual validation effort by ˜20–35% on recurring monthly deliverables (typical when checks are templated).

Selected Work

4 Production-Grade
AI Systems

Finance AI

FinanceIQ

Intelligent financial analysis platform that transforms raw financial data into actionable insights. Features automated spending pattern analysis, anomaly detection, budget recommendations, and beautiful visual dashboards for comprehensive financial intelligence.

LangGraphGroqPandasData AnalysisFastAPI

Open live app View on GitHub →

Travel AI

Deployed

TravelBuddy

Multilingual AI travel companion with GPS-powered location discovery, personalized itinerary recommendations, and persistent conversation memory. Features Groq + Gemini fallback for reliable responses, Google Places integration, and seamless travel planning experience.

FlaskGoogle PlacesGroqGeminiDocker

Open live app View on GitHub →

Multi-Agent · Coding

App Builder

Revolutionary natural language to production-ready code system. Multi-agent architecture with Planner, Architect, and Coder agents working together to simulate real software development workflows—from concept to deployment-ready applications.

LangGraphGroqPythonMulti-Agent

View on GitHub →

Research AI

Dr. Research

Advanced AI research assistant that autonomously conducts comprehensive research, analyzes multiple sources, synthesizes findings, and generates detailed research reports. Features intelligent source validation, citation management, and structured output formatting.

LangGraphReActResearchAI Agent

Open live app View on GitHub →

What I Build

AI Solutions for
Real Business

AI Agent Development

Custom autonomous agents using LangGraph — ReAct, self-reflection, task decomposition. From single tools to complex multi-agent orchestration with memory.

Workflow Automation

Replace manual processes with intelligent AI pipelines. Customer support, document processing, research workflows — running reliably without human intervention.

RAG & Knowledge Systems

Intelligent Q&A over your documents using vector embeddings and ChromaDB. Understands context, not just keywords.

Full-Stack AI Apps

React frontend + FastAPI backend + Docker. Real-time streaming, beautiful UIs, production-ready. Full product, not just an API wrapper.

Prompt Engineering

Structured architectures, JSON enforcement, hallucination reduction, guardrails. Making LLMs reliable and consistent in production.

ML & Data Analysis

Supervised/unsupervised models, feature engineering, intelligent dashboards. Raw data into actionable business intelligence.

What Clients Say

Honest feedback from
real projects

"I've worked with Kartikey on a few AI projects and he's genuinely good. Clean, well-structured code and a real understanding of agent architectures — not just API wrappers. Ships fast and is easy to build with."

Ayush Mittal ↗

Software Engineer (AI)

"Good experience overall. The research agent saves us time on desk research. We still tweak it sometimes for niche topics, but the base system is strong."

Sarah K.

Head of Research

"Understood our requirements quickly and shipped on the agreed timeline. Some edge cases came up later, but he helped us handle them without drama."

Aarav P.

Founder, AI Agency

"The finance tool is now part of our workflow. UI is clean, and streaming responses feel nice. There were a couple of minor bugs at launch, but they were fixed fast."

Priya S.

Product Manager

"Knows his way around production AI. Not just fancy demos. The RAG system is reliable enough that our team actually uses it daily."

Marcus L.

Senior Engineer

"Worked alongside Kartikey on agent and automation systems. He really knows his way around AI agents and harness engineering, and the work held up in production. Sharp, reliable, and someone I'd happily collaborate with again."

uddit ↗

AI Agents · Harness Engineering · Enterprise Automation & DevOps

Ayush Mittal ↗

Software Engineer (AI)

"Good experience overall. The research agent saves us time on desk research. We still tweak it sometimes for niche topics, but the base system is strong."

Sarah K.

Head of Research

"Understood our requirements quickly and shipped on the agreed timeline. Some edge cases came up later, but he helped us handle them without drama."

Aarav P.

Founder, AI Agency

"The finance tool is now part of our workflow. UI is clean, and streaming responses feel nice. There were a couple of minor bugs at launch, but they were fixed fast."

Priya S.

Product Manager

"Knows his way around production AI. Not just fancy demos. The RAG system is reliable enough that our team actually uses it daily."

Marcus L.

Senior Engineer

uddit ↗

AI Agents · Harness Engineering · Enterprise Automation & DevOps

FAQ

Frequently Asked
Questions

What services do you offer? +

I specialize in AI agent development using LangGraph, full-stack AI applications (React + FastAPI), RAG and knowledge systems, workflow automation, prompt engineering, and ML/data analysis. If it involves making AI reliable in production, that's my zone.

How much do you charge? +

Pricing depends on project scope. I offer hourly rates ($30–60/hr) or fixed-price project quotes. Most small-to-medium AI agent projects fall between $500–$2000. Email me and I'll send a detailed quote within 24 hours.

What does your development process look like? +

Discovery call → architecture design → iterative build with weekly demos → production delivery with Docker, docs, and deployment support. You always see progress, no surprises.

Can you work with my existing tech stack? +

Yes. I build standalone AI agents that integrate via APIs, or I work within your current stack. My FastAPI backends are modular and designed to plug into existing infrastructure easily.

Do you sign NDAs / keep projects confidential? +

Yes. I can sign an NDA, work under confidentiality clauses, and follow your security requirements. If needed, I can also work inside your GitHub org with limited access.

Do you provide post-launch support? +

Yes. I offer a handover with documentation, monitoring guidance, and a support window for bug fixes. Ongoing maintenance and feature iterations are available as a monthly retainer.

What do you need from me to get started? +

A quick brief (goals, users, success metrics), any existing docs/APIs, and sample inputs/outputs. After a short discovery call, I’ll share a clear scope, timeline, and quote.

How long does a project typically take? +

Simple AI integrations: 1–2 weeks. Single agent tools: 2–3 weeks. Multi-agent systems or full-stack apps: 3–6 weeks. I give accurate timelines upfront and I hit them.

Building

Autonomous

AI

Systems

That

Actually

Ship

A little
about me

Where I've
Built & Contributed

4 Production-Grade
AI Systems

AI Solutions for
Real Business

Full Tech
Stack

Idea to
Deployed Product

Honest feedback from
real projects

Frequently Asked
Questions

Building Autonomous AI Systems That Actually Ship

A littleabout me

Where I'veBuilt & Contributed

4 Production-GradeAI Systems

AI Solutions forReal Business

Full TechStack

Idea toDeployed Product

Honest feedback fromreal projects

Frequently AskedQuestions

Building

Autonomous

AI

Systems

That

Actually

Ship

A little
about me

Where I've
Built & Contributed

4 Production-Grade
AI Systems

AI Solutions for
Real Business

Full Tech
Stack

Idea to
Deployed Product

Honest feedback from
real projects

Frequently Asked
Questions