vishal.singh
Contact
Available for opportunities · Bengaluru, India

Building AI systems
that ship, not just notebooks.

I'm Vishal Singh, an applied GenAI engineer focused on multi-agent systems, local & cloud RAG, and edge computer vision — built with Python, LangChain, YOLOv8, and Next.js.

Contact MeView Projects
Vishal Singh

Vishal Singh

AI / GenAI Engineer

zsh — vishal@portfolio
➜~
About

Applied GenAI, end to end

From perception pipelines to production frontends — I care about AI that works outside a notebook.

Computer Science graduate focused on applied GenAI — multi-agent systems, retrieval-augmented generation, and edge computer vision. Over the past year I've shipped a multi-agent AI discovery engine, a local-first RAG assistant, and a real-time computer vision pipeline for autonomous UAVs. I work across the stack: Python for AI/ML, LangChain and Llama/Mistral for reasoning, YOLOv8 and ROS2 for perception, and Next.js for the interfaces that put it all in front of people.

Languages

  • Python
  • TypeScript
  • SQL / PostgreSQL

AI / ML

  • PyTorch
  • Transformers
  • YOLOv8
  • Fine-tuning
  • Embeddings

Generative AI

  • LangChain
  • RAG
  • FAISS
  • Ollama
  • Llama 3
  • Mistral
  • Agents

Platforms

  • ROS2
  • AirSim
  • Supabase
  • Vercel
  • Hugging Face
  • Agentforce

Multi-Agent Systems

Autonomous agent pipelines that discover, verify, and act on real-world data with LLM-based cross-checks.

Local & Cloud RAG

Retrieval-augmented generation, from quantised local inference to low-latency cloud vector search.

Edge Computer Vision

Fine-tuned YOLOv8 perception pipelines running real-time object tracking and threat classification.

Client Work

Deployed to production

Paid engagements shipped end to end — scoped with the client, built around their constraints, and running on their infrastructure.

hitechpack.com
Live in productionClient engagement

Hi-Tech Pack

B2B Production Infrastructure Modernization

Industrial packaging machinery manufacturer · 15+ machine categories

Role
Independent Developer / Forward Deployed Engineer
Categories
15 machine lines
Brand legacy
26 years
Runtime servers
Zero
Hosting
cPanel shared
Next.jsTypeScriptStatic ExportcPanel / Shared HostingB2B PlatformPerformance Optimization
Live Site

01/Context

Hi-Tech Pack required a complete modernization of their web presence to showcase 15 heavy machinery categories to international B2B clients, modernizing a 26-year legacy brand.

02/Constraint

The client required deployment on existing, restrictive cPanel shared hosting — without recurring cloud infrastructure or container maintenance costs.

03/Solution

Architected a decoupled static architecture using Next.js Static Exports (output: 'export'), custom asset pathing, and zero runtime server dependencies. Handled routing, responsive machine grids, and optimized client build delivery under tight hosting constraints.

04/Impact

Shipped a production-grade, ultra-low latency, zero-maintenance digital platform directly to the client's live environment at hitechpack.com.

Selected Work

Things I've shipped

Production systems and simulations spanning multi-agent orchestration, RAG, and computer vision.

AI Tool Hunter
hover to play

AI Tool Hunter

Autonomous multi-agent AI discovery engine

Live

A production multi-agent platform that autonomously discovers, validates, categorizes, and continuously updates a database of AI tools. A live search agent (Tavily + Llama 3.3-70B) scans the web for tools not yet in the database and cross-references pricing and feature data with an LLM before any write — a hallucination-proof pipeline with zero stale entries. The dataset is exposed as both a public Hugging Face dataset and an MCP server, so it's usable by end users in a browser and by AI agents/IDEs directly.

Pipeline

Fully autonomous

Refresh

Daily cron

Integration

MCP server

Next.js 16TypeScriptSupabaseLlama 3.3-70BGroqTavily APIGitHub ActionsMCP
Live SiteGitHubHF Dataset
Autonomous Defence Drone Agent

Autonomous Defence Drone Agent

Real-time UAV surveillance & threat intelligence

Exhibited

An autonomous UAV surveillance pipeline built on AirSim and Gazebo, integrating computer vision, multi-object tracking, threat classification, and decision intelligence for real-time reconnaissance. YOLOv8 was fine-tuned on custom defence datasets and wired into a modular perception → tracking → threat-classification → alert pipeline, with a live Streamlit command dashboard for telemetry, threat history, and mission logs. Selected for the KSCST State-Level Project Exhibition 2026.

Inference

15–25 FPS

Recognition

KSCST 2026

Pipeline

4-stage perception

PythonROS2YOLOv8PyTorchAirSimOpenCVStreamlitCUDA
GitHub
LLM PDF Research Assistant

LLM PDF Research Assistant

Local-first RAG, evolved into a cloud deployment

Live

A Retrieval-Augmented Generation system for querying PDFs in natural language. Started as a local-first, privacy-focused prototype running entirely offline on Ollama, FAISS, and quantised Mistral models for sub-second, zero-cloud-dependency document queries — then evolved into a cloud-deployed architecture (Flask on Render, Groq Llama 3.3, Google embeddings) for public accessibility.

Retrieval

FAISS vector search

Latency

Sub-second local

Phases

Local → Cloud

LangChainFAISSOllamaMistralSentence TransformersFlaskGroq API
Live SiteGitHub
Writing & Credentials

Where I publish & what's verified

Technical write-ups, open datasets, and certifications backing the work above.

Dev.to

Technical blog

Write-ups on GenAI systems, RAG architectures, and agent design.

Hugging Face

Datasets & models

Open-source datasets, including the AI Tools Corpus.

Credentials

  • KSCST State-Level Project Exhibition 2026

    Recognition · Selected for the autonomous UAV surveillance project (AIDF)

    View proof

  • Salesforce Certified Agentforce Specialist

    Salesforce · 2025 · 100% in Data Cloud integration — autonomous AI agents & enterprise automation

    View credential

  • Prompt Tuning for NLP

    Hugging Face · 2025 · Fine-tuning transformer models for NLP applications

  • Generative AI Fundamentals & Prompt Engineering

    IBM SkillsBuild · 2024

  • Qlik Sense Qualification Exam – Data Architect 2025

    Qlik · 2025

    View credential
Let's build something

Open to GenAI & applied ML roles

If you're building multi-agent systems, RAG pipelines, or vision-driven products, I'd like to hear about it.

Bengaluru, India
vishal.aisys@gmail.comResume