KISLAY/631
Say hello

Generative AI · Data systems · Human curiosity

I build useful intelligence for messy worlds.

I’m Kislay Kashyap — an engineer who likes hard problems, clean systems, and work with a pulse. This is part portfolio, part lab notebook, part self-portrait.

Enter the work Open résumé
Current mode: turning ambiguity into something shippable.
DATA LANGUAGE SYSTEMS CURIOSITY K/631 — 00.01
GENAI SYSTEMS✦RAG EVALUATION✦DATA ENGINEERING✦PRODUCT THINKING✦CLEAR WRITING✦
01

Not a résumé

I’m interested in the point where engineering becomes judgment — where a system stops being merely clever and starts becoming useful.

I like complexity only when it earns its place. I care about the last 10%: the part that makes a system reliable, explainable, and pleasant to use.

My best work happens between disciplines — engineering, product, editorial thinking, and the awkward real-world context that never fits neatly in a ticket.

01 / SIGNAL

Find the thing that matters.

More data is not automatically more insight. I design for signal, relevance, and decisions.

02 / CRAFT

Make the invisible feel obvious.

Good architecture, observability, and language should make a complex system easier to trust.

03 / MOTION

Ship, measure, sharpen.

Momentum matters. So does knowing when speed is creating debt instead of progress.

02 / SELECTED SYSTEMS

Work that had to work.

Selected chapters, not an exhaustive archive. The details are simplified; the outcomes are real.

01 — INTELLIGENCE AT SCALE ENTERPRISE GENAI

From noisy articles to executive signal.

Built production AI pipelines that ingest, cleanse, enrich, tag, summarize, and rank thousands of scientific and pharmaceutical stories into decision-ready intelligence.

DatabricksPySparkSpark SQLLLMsPython
1000sarticles processed daily
E2Eproduction intelligence flow
1 goalless noise, better judgment
02 — EVALUATION INFRASTRUCTURE RAG / RELIABILITY

Evaluation that makes RAG honest.

Designed a reusable evaluation framework across retrieval quality, response accuracy, latency, cost, and traceability — so teams could experiment faster without lowering the bar.

LangChainFAISSCI/CDPythonObservability
60%less manual evaluation
3×experimentation throughput
35%less debugging time
03 — INDEPENDENT BUILD LLM PRODUCT

A résumé that rewrites itself — responsibly.

Built an LLM-powered tailoring platform that parses source documents, adapts content to a role, and generates polished LaTeX PDFs with configurable similarity controls.

FlaskOpenAILaTeXAWS EC2S3
70% less manual tailoring effort Human judgment remains in the loop.
03 / THE PATH SO FAR

Career, but human.

Three roles. One recurring theme: use language, data, and software to make difficult knowledge easier to act on.

The conventional version
  1. 2025 — NOW

    Data Scientist · Generative AI

    LTIMindtree

    Enterprise intelligence systems, large-scale data pipelines, LLM workflows, performance, reliability, and the judgment layer between all of them.
  2. 2023 — 2025

    Software Engineer · Generative AI

    Persistent Systems

    RAG evaluation, traceability, automated experiments, backend services, AWS, and reusable foundations that made teams move faster.
  3. 2023

    Machine Learning Intern

    Josh Software

    NLP, BERT, GPT-3, vector search, summarization, and the first real proof that language systems could become a career.
04 / OPERATING SYSTEM

How my brain likes to work.

No productivity mythology. Just a few patterns that keep showing up.

↘

Messy inputs

Unstructured data. Vague business asks. Conflicting signals. Half-known constraints.

✳

Structured curiosity

Ask sharper questions. Prototype the smallest truth. Measure. Simplify. Repeat.

↗

Useful systems

Reliable enough for production. Clear enough to explain. Human enough to adopt.

Clarity is kindness.

Quality compounds.

Curiosity needs deadlines.

Not building is also a decision.

05 / LIVE-ISH STUDIO FEED

A page should feel alive.

This is a playful, curated simulation of the kinds of things usually moving through my head — not a surveillance feed, not a productivity score.

BUILD THINK TUNE SHIP
KISLAY / STUDIO_STREAM
--:--:--
stream active local time
› watching for the next useful signal
06 / TOOLKIT

Tools are temporary.
Good judgment travels.

Python OpenAI Databricks FastAPI PySpark LangChain FAISS AWS Docker Spark SQL Hugging Face Flask GitHub Actions Pandas Scikit-learn
99.2CAT percentile
8.7CSE CGPA
∞questions still open
07

Open channel

Let’s build something that deserves to exist.

Good conversations are welcome — especially around GenAI, data systems, product-minded engineering, or an unusually difficult problem.

kislaykashyap631@gmail.com
LinkedIn ↗ LeetCode ↗ Résumé ↗