You will get An audit of your production RAG and a tested fix PR in 3 days
Rising Talent
Project details
Is your RAG pipeline retrieving the wrong chunks, hallucinating, or quietly degrading over time?
I will audit your RAG/LLM pipeline and deliver:
1. A concise production-readiness report covering:
• retrieval quality on your real queries
• likely failure modes: hallucination patterns, retrieval drift, embedding mismatches, hybrid-search issues
• missing evals, validation gaps, and observability problems
• prioritized fixes with effort estimates
• code-level recommendations with file references where repo access is provided
2. One small, scoped fix PR from the priority list, with tests where the repo setup supports it.
Relevant experience: Postgres + pgvector, Pinecone, Weaviate, LangChain, LlamaIndex, hybrid search, RRF, reranking, and custom retrieval pipelines.
Recent open-source work: submitted a PR to garrtyan/gbrain adding tests for silent NaN similarity failures in a RAG pipeline.
If your AI feature works in the demo but feels unreliable in production, this audit is built to find the weak points.
I will audit your RAG/LLM pipeline and deliver:
1. A concise production-readiness report covering:
• retrieval quality on your real queries
• likely failure modes: hallucination patterns, retrieval drift, embedding mismatches, hybrid-search issues
• missing evals, validation gaps, and observability problems
• prioritized fixes with effort estimates
• code-level recommendations with file references where repo access is provided
2. One small, scoped fix PR from the priority list, with tests where the repo setup supports it.
Relevant experience: Postgres + pgvector, Pinecone, Weaviate, LangChain, LlamaIndex, hybrid search, RRF, reranking, and custom retrieval pipelines.
Recent open-source work: submitted a PR to garrtyan/gbrain adding tests for silent NaN similarity failures in a RAG pipeline.
If your AI feature works in the demo but feels unreliable in production, this audit is built to find the weak points.
AI Development Type
Knowledge Representation, Software MaintenanceAI Development Language
PythonWhat's included $249
These options are included with the project scope.
$249
- Delivery Time 3 days
- Number of Revisions 1
- Source Code
Frequently asked questions
About Nika
Next.js & Supabase Developer | Vercel Deploys | Stripe & AI App Fixes
Tbilisi, Georgia - 10:23 pm local time
I work in your stack: Next.js, Supabase, Vercel, OpenAI and Claude. I start from your repo and the exact breakage, not a rebuild.
Built it with Lovable, Bolt, v0, Cursor, or Replit? The tool layer rarely matters. Underneath it is Next.js, Supabase, and Vercel, my daily stack.
Common fixes:
- Users logged out on refresh (Supabase auth, @supabase/ssr, session flows)
- OAuth redirect and login loops
- RLS policies blocking real users
- Vercel build, deploy, and env var failures
- Stripe webhook failures
- OpenAI, Claude, and Gemini integration failures
How I work: you get a fixed quote on a small scope before any work starts. The fix ships as a pull request with a short note on what was wrong. I fix what you have, no rebuild pitch.
Proof: my portfolio includes a live Next.js 14 + Supabase marketplace for Georgian local food producers. It uses RLS policies, auth flows, and production deploys on Vercel, the same stack this profile is about. When your app hits a Supabase auth error, a failing deploy, or a Next.js bug, I bring the same production-debugging patterns to your repo.
I also debug LLM integrations (OpenAI, Claude, Gemini).
Send the repo link or the exact error message and where it appears, and I'll tell you what I'd check first before you commit to anything.
Steps for completing your project
After purchasing the project, send requirements so Nika can start the project.
Delivery time starts when Nika receives requirements from you.
Nika works on your project following the steps below.
Revisions may occur after the delivery date.
Day 1
I review your repo, RAG pipeline, sample queries, and expected outputs. I confirm the audit scope, identify likely failure points, and start the technical analysis.
Day 2
I share preliminary findings, rank the issues by impact and effort, and confirm one small, scoped fix to implement where repo access and setup allow it.