Available for AI Engineering Contracts

I Build
|

I don't just wrap APIs — I architect scalable agent swarms like Quest (10-Agent System) that generate 30-page deep-dive reports in under 2 minutes. From low-latency Voice Agents to quantized VLMs on the edge, I build and ship production-grade AI.

10+

Agent Swarms

<2 min

30-Page Reports

95%

SQL Accuracy

25%

RAG Improvement

View My Work ↓ Get In Touch

LangGraph Gemini Live RAG / Hybrid Search Qwen2.5-VL Fine-Tuning (Triplet Loss) Quantization (bf16 / GGUF) FastAPI / WebSockets Docker GCP / Azure PostgreSQL (pgvector) Prompt Engineering Context Engineering Llama 3 CI/CD Pipelines Microservices Real-Time Audio/Video LangGraph Gemini Live RAG / Hybrid Search Qwen2.5-VL Fine-Tuning (Triplet Loss) Quantization (bf16 / GGUF) FastAPI / WebSockets Docker GCP / Azure PostgreSQL (pgvector) Prompt Engineering Context Engineering Llama 3 CI/CD Pipelines Microservices Real-Time Audio/Video

Experience

Where I've Shipped Production AI

AI Engineer

Fraterny — fraterny.com/quest

July 2025 – Present · Remote

10-Agent Swarm Architecture: Architected "Quest," a distributed Multi-Agent System (LangGraph) that orchestrates 10 specialized agents to generate 30-page psychological reports in under 2 minutes.
Zero-Hallucination RAG: Fine-tuned the all-mpnet-base-v2 embedding model using Contrastive Learning (Triplet Loss), improving semantic retrieval accuracy by 25% for clinical domains.
State Consistency: Engineered a Long-Context Memory layer combined with Hybrid Search, ensuring 100% output consistency across complex, multi-turn user sessions.
High-Scale Async Backend: Implemented asynchronous processing (asyncio) in FastAPI to handle intensive reasoning tasks (140s+ context windows) without blocking concurrent users.
Cloud Infrastructure: Managed a Dual-Tier deployment on GCP, optimizing costs by routing traffic between lightweight sequential agents and heavy parallel swarms.

Projects

Systems I've Engineered

🧠

Quest

Multi-Agent Swarm · LangGraph

A distributed 10-agent system that orchestrates specialized AI agents to generate comprehensive 30-page psychological deep-dive reports in under 2 minutes with zero hallucinations via fine-tuned embeddings and hybrid search.

LangGraph FastAPI GCP RAG Triplet Loss

10 Agents

<2 min Generation

25% RAG Boost

🏋️

AI Fit Trainer

Real-Time Multimodal Agent · Gemini Live

A low-latency mobile fitness trainer using FastAPI WebSockets and Gemini Live Native Audio for sub-second voice interaction, real-time form correction, and bandwidth-optimized video streaming on minimal connections.

FastAPI Gemini Live WebSockets Context Eng.

Sub-sec Voice

Real-Time Video

Low-BW Optimized

🔍

Local Lens

Privacy-First Edge AI · Qwen2.5-VL

A local semantic image search engine powered by Qwen2.5-VL-3B quantized to bf16, running entirely on consumer hardware (24 GB VRAM) without cloud dependency. Sub-second retrieval via FAISS indexing.

Qwen2.5-VL Quantization FAISS Python

3B Params

bf16 Quantized

Sub-sec Search

⚡

Enterprise Text-to-SQL

Tesla ERP Case Study · LangChain

A high-precision Text-to-SQL Agent translating complex natural-language queries into optimized SQL with 95% syntax accuracy. Dynamic Schema Retrieval and strict prompt guardrails eliminate hallucinations entirely.

LangChain Text-to-SQL Multi-Agent ERP

95% Accuracy

100% Schema Adherence

Education

Academic Foundation

Christ University

M.Sc. Artificial Intelligence & Machine Learning

Online · Expected 2027

Guru Gobind Singh Indraprastha University

B.C.A. (Specialization in AI/ML)

New Delhi, India · Graduated 2025

Certifications

🤗

Quantization Fundamentals & Preprocessing Unstructured Data

Hugging Face

🔬

Introduction to AI

IBM

🎓

Supervised Machine Learning: Regression and Classification

Stanford Online (Coursera)

🐍

Data Science and Machine Learning on Python

Python Institute

☁️

Introduction to Amazon SageMaker

AWS

Contact

Let's Build Something Extraordinary

Get in Touch

Have a project that needs production-grade AI? Whether it's a multi-agent architecture, a real-time multimodal system, or an edge AI deployment — I'm ready to ship it.

✉️ harsimransingh27448@gmail.com 📞 +91 7814351011 🔗 linkedin.com/in/harsimransinghtech 💻 github.com/harsimran726

I Build |