hemant01
Open to freelance projects

HEMANT KUMAR / AI & SOFTWARE DEVELOPER

Small screen.
Big ideas.

A little about me — this obsession with creating things started with origami, then moved to painting, and eventually led me to software development. Since then, I’ve built software across different domains and explored a wide range of technologies. Feel free to go through my projects and see for yourself whether I can build or not

Explore my work ↗Let’s build something ↗
BASED IN INDIAWORKING EVERYWHERE
HK / CREATIVE HARDWAREON
POCKET OS 2.0HK
Hemant KumarYour next
good idea.
● EXPLORE MODE
COLOR DISPLAY / LIMITLESS POSSIBILITIES
pocketfolioCOLOR EDITION — 02
BACK
SELECT
SELECT
START
BUILT FOR CURIOSITY●
9 PROJECTS / ONE CURIOUS DEVELOPERSCROLL TO EXPLORE ↓
01 /THE PERSON BEHIND THE PIXELS

Good software
starts with
a good question.

I care about what happens after the demo: clear answers, reliable workflows, and a product that people can actually use.

HOW I THINK

If we can imagine it, we can build it. The only question is: how?

01Understand deeply.ASK / LISTEN / MAP

Start with the people, the workflow, and the problem worth solving.

02Build deliberately.DESIGN / CONNECT / TEST

Give each part a clear job. Keep the interface simple and the system maintainable.

03Make it useful.LEARN / REFINE / REPEAT

Test real inputs, examine the failures, and improve the parts that matter.

02 /EXPERIENCE

Built through
doing.

From an idea on paper
to a system that works.

EXPERIENCE / OPEN THE WORKBENCH01 CHAPTERS
01India / WORK LOG
Sep 2025 – Present

Python Developer

Prishitech Solutions

Building practical AI across logistics, enterprise support, industrial systems, and mobile applications.

8 FOCUS AREAS9 CORE TOOLS
Explore this chapterInside the workResponsibilities, tools & approach
  1. 01

    Built a pre-LLM rule layer to identify SAP module context, retrieved the knowledge base from ChromaDB, and used LLaMA 3.2 through Ollama to draft resolutions. An IMAP/SMTP pipeline connects the agent to email.

  2. 02

    Built a Qwen agent to analyse route data and flag alternatives for logistics decisions.

  3. 03

    Built the system from scratch with Playwright to handle rendering, session tokens, and request headers. Added Qwen to turn unstructured carrier text into normalized summaries with predicted ETAs.

  4. 04

    Built a GPT-4 extraction pipeline; batched API calls and cached repeated field patterns to improve throughput.

  5. 05

    Set up a cron-based ingestion pipeline that downloads daily market files, parses their contents, and loads them into PostgreSQL.

  6. 06

    Designed a fully offline platform using TF-IDF with Logistic Regression and Random Forest classifiers to return behavioral insights.

  7. 07

    Built an Agentic AI system that reads sensor data, classifies the fault type, and automatically starts a resolution workflow.

  8. 08

    Built an Android app backed by FastAPI, using LLaMA 3.2 through Ollama. Quantized the model and shortened prompts to reduce inference latency.

ON THE WORKBENCH
PythonQwenLLaMA 3.2GPT-4OllamaPlaywrightChromaDBFastAPIPostgreSQL

From roughly 60 to 85 documents per hour: batching and caching made a measurable difference in the document extraction pipeline.

DECEMBER 2025 / BONUS CHAPTER

5 hours.
One build.
Third place.

Developed and presented a prototype addressing a real-world domain problem using Generative AI tools within a 5-hour sprint, securing 3rd place.

Tap the badge to open the story.
THE SMALL CHANGES ADD UP

What does 60 → 85
look like over time?

Explore the document pipeline’s reported hourly throughput.

Before
~60
After
~85
~25 more documentsIllustrative projection at constant throughput, not an additional measured benchmark.
03 /SELECTED WORK

Ideas, made
real.

A closer look at the problem,
the process, and the product.

SELECTED FILES (09)EXPLORE THE COLLECTION ↓

9 projects

01AI / ENTERPRISE
AI / ENTERPRISE

SAP MM Support Agent

↗

An inbox full of SAP issues. A local agent with context.

PythonLLaMA 3.2OllamaChromaDBRAGIMAP / SMTP
02AI / LOGISTICS
AI / LOGISTICS

Shipment Optimization

↗

Better routes begin with better questions.

PythonQwenOllamaSQL
03DATA / AUTOMATION
DATA / AUTOMATION

Container Tracking

↗

Carrier chaos, translated into a clear next step.

PythonPlaywrightQwenLLM Integration
04AI / AUTOMATION
AI / AUTOMATION

Document Extraction

↗

Less paperwork. More documents through the door.

PythonGPT-4REST APIsCaching
05DATA / AUTOMATION
DATA / AUTOMATION

Nightly Market Data

↗

The market closes. The pipeline clocks in.

PythonCron JobsPostgreSQLLinuxSQL
06AI / ML
↗

Useful intelligence. Even without a connection.

Pythonscikit-learnTF-IDFLogistic RegressionRandom ForestSentiment Analysis
07AI / INDUSTRIAL
↗

From a sensor signal to a resolution workflow.

PythonAgentic AISensor dataClassification
08AI / MOBILE
↗

An outfit, decoded in under two seconds.

PythonFastAPILLaMA 3.2OllamaQuantization
09MOBILE / PRODUCT
LIVE ON GOOGLE PLAY

MANTIS

↗

A React Native study tracker, available on Google Play.

React NativeAndroid
Get it on Google Play ↗

Different problems.
The same care in the details.

IDEA → SYSTEM → EXPERIENCE
THE WAY I BUILDKEEP SCROLLING ↓

MAKE IT
MEAN SOMETHING.

Thoughtful systems. Useful software. People first.

FROM A SMALL IDEATO WHAT’S
NEXT.Explore the toolkit ↗
INTENT → CRAFT → IMPACTSCROLL TO TRANSFORM
04 /THE TOOLKIT

Built on
good foundations.

TOOLS FOR THE TASK AT HAND⌘
TOOLKIT / 01

The building blocks.

PythonC++

From automation scripts to complete AI systems.

TOOLS CHANGE. GOOD THINKING LASTS.
A TOOL IS ONLY AS GOOD AS ITS FIT

Give the toolkit a mission.

The unplugged toolkit.

Local models, retrieval, and classifiers for work that stays on client hardware.

PythonOllamaLLaMAChromaDBTF-IDFscikit-learn
05 /WAYS WE CAN WORK TOGETHER

Your next problem.
My kind of project.

01AI systemsRAG & document assistants

Give your team a useful way to work with its knowledge.

  • RAG & document assistants
  • LLM integrations
  • AI workflow prototypes
Python / LLMs / vector search
Let’s build this ↗
02Software that fitsWeb applications

Turn a specific business problem into a focused digital tool.

  • Web applications
  • Backend APIs
  • Internal tools
React / Next.js / FastAPI
Let’s build this ↗
03Less busyworkData pipelines

Connect your data and take repetitive tasks off the list.

  • Data pipelines
  • Browser automation
  • Reporting workflows
Python / SQL / Playwright
Let’s build this ↗
OPEN FOR COLLABORATION

Have something
worth building?

Bring the messy problem, the ambitious idea,
or the thing you wish worked better.

Let’s make it happen ↗
06 /THE NEXT CHAPTER

Start with
a conversation.

No perfect brief required.
Just tell me what you have in mind.

Project inquiry
Optional · up to 3 MB
Your details are used only to reply
to your project inquiry.