bobross-ui · GitHub
Skip to content
View bobross-ui's full-sized avatar

Block or report bobross-ui

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
bobross-ui/README.md

Kshitij Ghode

Full-stack engineer building AI systems that hold up past the demo: agent memory, grounded retrieval, and real-time apps with real users. I ship full products end-to-end, from the database schema to the screen.

Currently open to backend / AI-infrastructure roles.

Featured Work

CAT-DUEL - real-time 1v1 exam-prep duels

Two players, one question set, live Elo matchmaking, and timed scoring. The UI is the easy part; the hard part is keeping game state consistent when both players answer on the same millisecond and connections drop mid-duel. Solved with two-phase locking and Redis NX guards for race-free concurrent scoring, plus reconnect-safe state so a dropped player rejoins exactly where they left off.

TypeScript | React Native | Expo | Node.js | Socket.io | Postgres | Redis | Prisma

brain - local-first memory layer for long-running agents

Scoped memories over raw conversation history with hybrid vector + BM25 retrieval, automatic fact extraction, and reconciliation when new facts contradict old ones, all exposed as MCP tools so any agent can plug in. Built with an eval harness from day one, so retrieval quality is measured, not guessed.

LOCOMO conv-26 reranker blend: 64.8% judged accuracy (129/199), 73.6% fused evidence recall, 77.7% evidence hit rate, 0 failures. Single-conversation experiment, 199 questions.

Python | SQLite | sqlite-vec | FTS5 | Ollama | MCP

cortex - social exports into a grounded knowledge base

Ingests personal exports into a pgvector store and answers questions with cited sources, so every claim traces back to where it came from. Hybrid retrieval with reciprocal rank fusion instead of naive score addition.

Python | FastAPI | Postgres | pgvector | React | DeepSeek

What I'm Into

AI that's more than a chat box: memory that gets better the longer it runs, retrieval that can show its work, and real-time interfaces that make complex flows feel direct. Most of what I build is solo, in public, and shaped by real users instead of a slide.

Stack

Languages | TypeScript | Python | SQL | JavaScript

Frontend | React | Next.js | React Native | Expo | Tailwind | Framer Motion

Backend | Node.js | Express | FastAPI | Prisma | PostgreSQL | Redis | SQLite

AI / Retrieval | RAG | embeddings | vector search | BM25/FTS | MCP | local LLM tooling

Infra | Docker | GitHub Actions | CI | testing | observability-minded design


Pinned Loading

  1. CAT-DUEL CAT-DUEL Public

    TypeScript 1

  2. janavani janavani Public

    Python

  3. devflow devflow Public

    Java

  4. cortex cortex Public

    Python

  5. brain brain Public

    Agentic memory layer

    Python

  6. notes notes Public

    TypeScript