§00
Pune, IndiaAvailable 2026 

LaxmiKhengare

AI engineer in Pune. I build agentic systems that do real work, then measure whether they actually did.

the measuring part is the fun part
Scroll
§01

Now

Laxmi Khengare, photographed outdoors against foliage.
Laxmi, Pune
A hand-drawn loop: build, then ship, then measure, then build again.BuildShipMeasure
the loop I care about

I build end-to-end AI systems in Python and FastAPI: retrieval, agent routing, evaluation and observability.

  • Graduated in AI and Data Science from Vishwakarma University this year, with three internships already behind me: recruitment automation at ZeepleWork, a renewable-energy assistant at SanchAI, and AI plus full-stack work at Millennium Techlink.
  • What I care about is the part most demos skip. Anyone can get a model to produce output; proving it is right needs a labelled test set, a number, and a regression suite that catches it getting worse.
  • Occam routes work across agents under a hard rupee budget and reports what that choice cost against a frontier-model baseline. It has never breached a budget in 701 recorded runs.
  • Clause answers questions about insurance policies with the source clause attached, and refuses when the policy does not say. Knowing when to refuse was the harder half.
Currently
AI & Fullstack Developer at Millennium Techlink
Working on
agent orchestration, retrieval, evals in CI
Open to
AI engineering roles in Pune or remote
§02

Work

  1. Apr 2026 to now

    Millennium Techlink

    Pune
    AI & Fullstack Developer
    • Build and integrate AI-powered features into a product-certification and automation platform, across model integration, backend services and UI.
    • Develop scalable web applications in Next.js, React, FastAPI and PostgreSQL, deployed with Docker and Git-based workflows.
    • Python
    • Next.js
    • FastAPI
    • PostgreSQL
    • Docker
  2. Jan to Apr 2026

    Sanchai Analytics

    Pune
    Agentic AI & Fullstack Developer Intern
    • Developed features for SanchAI, an agentic assistant for renewable-energy plants, surfacing performance insights for solar and wind assets.
    • Built LLM-driven workflows over plant operational data so engineers could query asset performance in natural language instead of reading reports.
    • Python
    • Agents
    • Time-series
    • React
  3. Sept to Dec 2025

    ZeepleWork Technologies

    Pune
    Agentic AI Developer Intern
    • Built agentic recruitment systems in Python and FastAPI that automated candidate sourcing across LinkedIn, GitHub and X-Ray search.
    • Designed multi-step agent workflows with tool calling and structured output parsing to screen and rank candidates against role criteria.
    • Python
    • FastAPI
    • Agents
    • Scraping
    started here with sourcing. still thinking about the screening half.
§03

Field Notes

  • P01
    Ask a plain-English question about a health insurance policy and get a sentence-level answer with the exact source clause highlighted in the original PDF. It refuses to answer when the policy does not cover the question, which is the harder half of the problem.
    • Python
    • FastAPI
    • pgvector
    • Docling
    • Next.js
    Read the notesCode
    it says no when it should
  • P02
    Decomposes a task into a DAG of subtasks, routes each to the cheapest agent that can actually do it under a hard rupee budget, and verifies every output before synthesis. The bandit router hit 96% of the frontier-model baseline's success rate at 49% of its cost.
    • Python
    • FastAPI
    • Postgres
    • Redis
    • Langfuse
    Read the notesCode
    0 budget breaches in 701 runs
  • P03
    GPU owners sell model access over encrypted peer connections. An on-device confidence router answers easy prompts locally for free and escalates only the ambiguous ones to a paid peer. Model weights never move; only prompt bytes in and token bytes out cross the wire.
    • TypeScript
    • QVAC
    • Hyperswarm
    • Solidity
    • Electron
    Read the notesCode
    the weird one
  • P04
    Agent memories are extracted inside a hardware-secured enclave, encrypted under a wallet-derived key and stored as transferable on-chain assets. Owners grant and revoke read access to other agents, and every access is logged immutably.
    • TypeScript
    • Solidity
    • 0G
    • TEE
    • HNSW
    Read the notesCodeLive
  • P05
    When a keeper service shuts down, its address stays embedded in contract roles that nobody can export. Understudy finds every contract that still trusts it, rebuilds the jobs, shadow-runs them, and only revokes the old credential once a judge has verified the replacement.
    • TypeScript
    • Foundry
    • viem
    • MCP
    • Blockscout
    Read the notesCode
    revoke too early and money leaves
  • P06
    A contract is written once, but the regulation around the asset never stops moving. The Warden buys regulatory updates, reasons about them, proposes a compliance module, has it audited by a second agent, installs it live, and can roll a bad version back in one transaction.
    • Rust
    • Odra
    • Casper
    • TypeScript
    • x402
    Read the notesCode
  • P07

    Agent Credit Desk

    An agent runs a USDT treasury, underwrites loan applications on a 30-second heartbeat, and deploys an on-chain escrow per loan. Idle capital sits in Aave rather than doing nothing, and other agents can borrow through the API.
    • TypeScript
    • Express
    • Solidity
    • Prisma
    • Aave
    Code
  • P08

    Blood Cell Analysis

    A convolutional classifier reaching roughly 90% accuracy detecting abnormal blood cells from microscopy images, using image-processing pipelines for preparation. Published by Springer.
    • Python
    • CNN
    • OpenCV
    Paper
§04

Stack

Roughly how the pieces fit together. Hover or tap a node to see what connects to what.
A hand-drawn map of the stack: agentic systems built from LangGraph, LLMs and RAG, on Python, running on FastAPI, Docker and PostgreSQL.Agentic AILangGraphLLMsSLMsRAGPython
  • Languages
    • Python
    • TypeScript
    • JavaScript
    • SQL
  • AI / ML
    • Multi-Agent Systems
    • Prompt Engineering
    • Fine-Tuning / PEFT
    • NLP
    • Computer Vision (CNNs)
    • Predictive Modeling
    • MLOps
  • Frameworks
    • Next.js
    • React.js
    • Express.js
    • Tailwind CSS
    • PyTorch / TensorFlow
    • LangChain
  • Retrieval
    • pgvector
    • Hybrid Dense + Lexical
    • RRF Fusion
    • BGE Reranking
    • Docling
  • Data
    • MySQL
    • MongoDB
    • Redis
  • Tools
    • Git
    • Linux
    • VS Code
    • Langfuse
    • REST APIs
    • CI/CD
§05

Contact

Currently open to AI engineering roles, Pune or remote.

If you are building agents, retrieval systems, or anything where the hard part is knowing whether it worked, I would like to hear about it.
I reply to everything
Email
laxmiofficial1611@gmail.com
Elsewhere
GitHubLinkedInRésumé