AI/ML & Full-Stack Engineer

George Babakhanov

I build AI products end to end. Now I'm going deeper: inference, AI architectures and the math behind models. I don't know it all yet, and that is the work I want to do.

Open to startup roles in San Francisco

George Babakhanov

Selected work

Things worth a click
Co-founder · iOSKavaApr 2026 - PresentKava is a free café map app for iPhone that I co-founded with Keyon. It maps the cafés in your city and shows the ones you haven't tried.Read more →Full-stack developerLOCVMFeb 2026 - PresentHealthtech startup that matches physicians with locum jobs. I came in to rebuild the matching engine and stayed as the full-stack developer.Read more →AI/ML & full-stack engineerAvateqJan 2025 - PresentAI/ML & full-stack engineer at a broadcast RF company. I built the ERP that runs it, an RF analysis app featured in TV Tech, and on-prem AI.Read more →OSS contributorOpen sourceSep 2026 - PresentI like to fix the hard issues that other people run into, and I learn something new each time. I write about my bugs and lessons on my socials. Here are my contributions to big open-source projects.See contributions →

Work timeline

Now
Jan 2021

All Works

Open Source - vorssaint-utils ContributorSep 2026 - Present · 1 min read
I contribute to vorssaint-utils, a free and open-source macOS menu bar toolkit with 26,000+ stars on GitHub. 4 pull requests merged so far, in Swift.
Kava - Co-FounderApr 2026 - Present · 3 min read
Kava is a free café map app for iPhone that I co-founded with Keyon. It maps the cafés in your city and shows the ones you haven't tried.
LOCVM - Full-Stack DeveloperFeb 2026 - Present · 2 min read
Healthtech startup that matches physicians with locum jobs. I came in to rebuild the matching engine and stayed as the full-stack developer.
Avateq - AI/ML & Full-Stack EngineerJan 2025 - Present · 3 min read
AI/ML & full-stack engineer at a broadcast RF company. I built the ERP that runs it, an RF analysis app featured in TV Tech, and on-prem AI.
Independent Developer & AI ConsultantJan 2025 - Present · 1 min read
40+ projects since January 2025 for clients and for myself -> AI systems, automation and websites. Some highlights below.
Sendable - Co-Founder & CTOMar 2026 - Sep 2026 · 1 min read
Email deliverability tool for people whose business runs on email. I was the only engineer. Paused. I plan to open-source the core as an agent skill.
AVATalks - AI Content System (via Riipen)Dec 2025 - Feb 2026 · 1 min read
Client project through Riipen. An n8n system that researches topics and writes LinkedIn and X posts for the client's ideal customer. The first posts got 80+ likes each vs the client's usual 4-10.
n8ncontent generationautomationLinkedInsocial media
PPF ResearchNov 2024 - Apr 2025 · 2 min read
Built Precision Propagation Filtering-an adaptive retrieval layer for RAG systems integrating a linear regression relevance estimator and neural inference layer for dynamic information expansion.
GlobalGym - Co-founderJun 2024 - Mar 2025 · 1 min read
Co-founded a community and fitness coaching platform focused on building genuine connections through health and wellness.
DucensAI - CTOJan 2024 - Apr 2026 · 1 min read
CTO of a small AI automation company. I built AI invoicing for a B2B hardware client, a private legal research assistant, and a security automation proof of concept.
Foodie - Co-founderJan 2023 - Apr 2023 · 1 min read
Built a unified restaurant loyalty platform for Toronto. Signed 56 restaurants including fast food and food chains like MrZagros Canada before launch.
CityScapes - Co-founderSep 2022 - Jan 2023 · 1 min read
Co-founded a company creating 3D printed city models as art pieces, turning urban landscapes into collectible miniatures.
DisClub - Co-founderJan 2021 - Dec 2023 · 1 min read
A community-focused app connecting like-minded people for friendships. Like Tinder, but for building genuine connections based on shared hobbies and interests.

PPF Research

Related code on GitHub (opens in a new tab)

George Babakhanov
George Babakhanov
Nov 2024 - Apr 2025 · 2 min read

Developed Precision Propagation Filtering (PPF), a method for improving retrieval precision in Retrieval-Augmented Generation (RAG) architectures, integrating a linear regression relevance estimator and neural inference layer for dynamic information expansion.

The Problem

Most RAG pipelines suffer from retrieval errors between 15–20%. These occur when too much irrelevant context overwhelms the model, or too little context leaves it guessing. GraphRAG's entity-relationship approach helps in some cases but can underperform when the knowledge graph doesn't align with query intent.

How PPF Works

PPF adds an adaptive filtering layer between the retriever and LLM:

Predictive Precision Model - A lightweight model predicts how much precision the retrieval step should apply. Some queries need breadth (multiple chunks), others need laser precision (1–2 chunks).

Dynamic Context Control - Adjusts retrieval dynamically to prevent "context flooding" (too many tokens) or "context starvation" (not enough). Uses a logarithmic cap function to keep context growth controlled even in massive vector DBs.

Reranking Layer - Integrates with rerankers like Cohere Rerank to ensure only the highest-quality chunks reach the LLM.

Technical Details

  • Target hallucination rate: <2%
  • Optimal performance with 256-token chunks
  • Model-agnostic and vector DB agnostic (Neo4j, Pinecone, Weaviate, Qdrant, FAISS)
  • Plug-and-play with existing ingestion pipelines

Status

Tested on Neo4j GraphRAG. The next step was standard RAG with cohesive hierarchical segmentation.

If you want to check it out, the code is on GitHub.

More in Work

All work →
George Babakhanov
Written by
George Babakhanov
I'm George (Gosha), an AI and full-stack engineer in Toronto. I ship production software -> the ERP that runs Avateq end to end, email and matching systems at LOCVM, and Kava, an iOS app I co-founded. I started building companies at 16 and I learn by building, breaking and fixing. Next -> a startup team in San Francisco.