# Jayesh Ranjan Kesari

AI Engineer · Agentic Systems & LLM Infrastructure

jrkesari@gmail.com · https://github.com/Jrkesari · https://www.linkedin.com/in/jrkesari · https://x.com/me_jayeshrkesar · https://jayesh.cv

## Profile

AI engineer building production agentic systems since 2024. I work on the LLM platform behind a multi-tenant enterprise product: model routing across providers, an MCP tool server, RAG over documents and spreadsheets, and token-level cost tracking.

## Skills

- **Languages:** Python, SQL, TypeScript, Go
- **Agentic:** MCP (client & server), tool-calling loops, LangGraph, LangChain, CrewAI
- **LLM:** Azure OpenAI, Anthropic Claude, Google Gemini, NVIDIA NIM; prompt caching, structured outputs, RAG, vision
- **Backend:** FastAPI, asyncio, Pydantic, gRPC, REST
- **Data:** DuckDB, SQLite + sqlite-vec, ClickHouse, Redpanda, FAISS, pandas
- **Platform:** Docker, Azure, Microsoft Graph, OpenTelemetry, Git, pytest

## Experience

### Netision Technology LLP · AI Engineer

May 2025 – present · Noida · on-site

**Agent runtime and tools**

- Wrote the agent loop that runs every AI query: tool schemas, capped iterations, result truncation, and a last tool-free call so the model has to answer in text.
- Set up a tag-based tool contract where tools declare what they do and what they need, so adding one takes no runtime changes.
- Removed an extra billed round on capped turns by offering file tools only when the prompt actually asks for a file.
- Added cooperative cancellation across the pipelines after finding that stopped queries were still writing their results.

**Multi-provider model layer**

- Built a shared library and model registry: 17 models across 6 providers, picked per request instead of baked into the pipeline; 86 releases in three months.
- Got prompt caching working with ordered breakpoints, bringing repeat tool-loop rounds down to a fraction of the first round's cost.
- Tracked down a content-filter block that only showed up through the proxy, then confirmed the fix on live calls: blocked calls went from 3/12 to 0/12.

**Token accounting**

- Built a per-user cost ledger that meters every paid call (chat, embeddings, vision, OCR, images) into SQLite with a ClickHouse mirror.
- Made missing usage and unpriced models record as such rather than as zero, so nothing free gets confused with something never measured.
- Read token counts straight from provider payloads so cached-input and reasoning tokens are not lost.

**Documents and spreadsheets**

- Routed uploads on text density rather than file extension, so image-only decks go to vision transcription and text PDFs to the normal extractor.
- Swapped naive chunking of spreadsheets for SQL over detected tables with schema-card embeddings, then made table discovery progressive so it fits the tool-result cap. Graded accuracy went from 2/8 to 6/8.
- Sandboxed model-written SQL: no external access, read-only statements, scoped to the current conversation's tables.

**Connectors, tools and security**

- Architected the MCP tool server: 49 tools, including 13 per-user connectors for Microsoft 365, Power BI, Google Workspace, Zoho and Databricks (the last two not yet run against a live account), passing delegated tokens as transport headers so the model can never set one.
- Designed host-side approval for outbound mail: the parked call is the card, and Send runs it with a fresh token and no model in the path. A teammate built the first version; I added replies, attachments and per-provider routing.
- Wrote most of 60+ test modules covering wire contracts, cancellation, metering and citations.

**Earlier at Netision**

- Text-to-SQL analytics and root-cause agents over a read-only warehouse; cut a root-cause turn from 77.8 s to 12 s with the report unchanged.
- Backends for an on-prem observability platform in Go and FastAPI: agents, a collector, a probe, and a control plane on Redpanda and ClickHouse.

### Coginetics · AI Software Trainee

Dec 2024 – May 2025 · Noida

- Built LLM agents with LangChain, CrewAI and the OpenAI API, where I moved from web development to agents.

### Tech Freedom Online · Web Developer Intern

Jul 2024 – Dec 2024 · Remote

- Front-end and back-end work on web applications.

## Projects

- **Intelligent web scraper with MCP.** An MCP server that exposes web scraping to any MCP client, working out the site type through parallel async probing; a multi-strategy scraper built on Factory and Strategy patterns. (Python, Playwright, BeautifulSoup, MCP)
- **Multi-agent automation suite.** 25+ CrewAI agents with sequential processing, shared memory and delegation for lead generation, SEO and content workflows. (CrewAI, Gemini, LangChain, pandas)
- **SplitCash: voice-driven finance pipeline.** Turns Hindi speech into translated, structured ledger entries, with 5 speech-to-text and 3 LLM providers behind one interface. (LangChain, DuckDB, Python)

## Education

- Bennett University · BCA, Data Science, 9.6/10 (2022 – 2025)

## Certifications

- Introduction to Model Context Protocol, Anthropic (2025)
- AI Fluency: Framework & Foundations, Anthropic (2025)
- Google Data Analytics Professional Certificate
- AWS Academy Graduate: Data Engineering, Cloud Architecting, Cloud Foundations