Skip to content
View MaithreshVaddi-27's full-sized avatar

Block or report MaithreshVaddi-27

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. 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
MaithreshVaddi-27/README.md

● SYS_STATUS: NOMINAL // AUTH: ED25519 // 10 SOLO SYSTEMS // 13 AUTOMATIONS // LOC: HYDERABAD [17.3850° N, 78.4867° E] // LATENCY: 12ms // ⌘K CONSOLE

$ whoami --role // maithresh.sh

Hi, I'm Maithresh Vaddi 👋

Final-year B.Tech CSE @ KMIT Hyderabad (CGPA 7.96/10) · building agentic RAG that works offline and survives production

Terminal session — whoami, mission, proof

Portfolio — maithresh.sh LinkedIn Gmail LeetCode HackerRank

maithresh.sh boot console — Maithresh Vaddi, AI/ML Engineer

$ ./metrics.sh --live — 10 solo systems, 13 automations, 100% offline inference


About · $ cat about.md

No inflated claims — every line here is something I can defend in an interview. Final-year B.Tech CSE @ KMIT Hyderabad (CGPA 7.96/10), building agentic RAG outside class: hybrid retrieval, MCP tool orchestration, LangGraph recovery loops. Now hardening TrustRAG verification + pushing CareerOS-Pro to Docker prod.

$ whoami — handle, role, base, focus, proof

$ cat about.md — defense pledge

$ curl stack.local/llm — local inference runtime telemetry


Stack · $ ls stack/ --grouped

$ ls stack/ --grouped — categorized production stack on terminal card

$ ls stack/ --grouped — full stack detail
  • llm.sh — LangChain · LangGraph (self-healing loops, multi-provider routing) · CrewAI · MCP server + client (JSON-RPC 2.0, Composio) · RAG (dense + BM25 + RRF) · cross-encoder rerank · NLI verify · Qdrant · ChromaDB · ONNX · bge-small/base, mpnet · Gemini · NVIDIA NIM · Ollama · llama.cpp · MLX · ReAct/Plan-Execute · memory/checkpointing · LoRA/RLHF. Method: directed AI-orchestrated engineering — architecture owned, implementation delegated.
  • automation.sh — n8n ×10 · Make.com · Automation Anywhere (RPA) · Webhooks · REST · Telegram Bot API · Groq Whisper · Redis.
  • backend.sh — Python · Java (OOP) · C++ · JS (ES6) · SQL · FastAPI · Granian · Flask · httpx · Pydantic · SQLAlchemy · Node/Express · MongoDB · MySQL · SQLite · Redis · Pandas/NumPy. 26 LeetCode SQL.
  • cloud.sh — Docker/Compose · AWS (labs) · Render · Cloudflare Pages · K8s · Jenkins · TF/Keras · sklearn · YOLO · Git · pytest/Vitest/Playwright/k6/Bandit/Trivy. Frontend (supporting): React 19, Framer Motion, Gradio, Bootstrap · Figma.

[01] Systems · $ ls projects/ --featured

TrustRAG — AI Reliability Workbench · flagship · FastAPI · React 18 · Qdrant · LangGraph · ONNX · MCP · Docker

route → hybrid retrieve → grounded generate → claim decompose + NLI (one fused call, −62% verify latency) → SHA-256 audit → score vs τ ≥ 0.75 → bounded recover (≤2) → answer / ABSTAIN

  • Offline by default: Ollama/llama.cpp/MLX + ONNX bge-small + int8 rerank. No API keys. Optional Gemini/NVIDIA NIM.
  • Perf work: KV-cache quant, flash attention, prompt caching, speculative decoding, adaptive top-k. JWT-HS256 + JTI denylist, SSRF-hardened URLs, Prometheus, A/B framework, MCP server.
  • 111 pytest + 15 Vitest + 2 Playwright E2E · k6 p95 < 300 ms · Bandit + Trivy CI gates.
Build One line
DocuChat Agentic PDF-RAG (CLI + Gradio): Gemini/Ollama/llama.cpp, Composio MCP routing, SQLite memory. Killed the 3-agent swarm → single agent: 6.8 s → 1.9 s, −68% tokens.
CareerOS-Pro 5 job APIs + async career-page scraper · deterministic normalize, 2-stage dedup (−74% redundant posts) · LangGraph fallback chain · HEAD+Firecrawl verify · /agents/health · Docker.
Resume Crew CrewAI resume↔JD matcher, evidence-only report (zero invented skills), 8-workflow Gradio + CLI, offline Ollama default, telemetry off.
Game Dev Crew Designer→Dev→QA pipeline, idea → playable WASM game (pygbag + ngrok). Fixed launcher-overwrite + async-loop freeze.

$ run workbench --select-system → live pipeline simulators + interview defenses on maithresh.sh.

$ ls agents/more/ — SkillMap · SalaryInsights · Inbox Summarizer · Blog Crew · CourseFinder
  • SkillMap — live India job search (Tavily+JSearch), checkpointed follow-ups.
  • SalaryInsights — Glassdoor/AmbitionBox/PayScale/Levels.fyi via Firecrawl MCP, sourced answers.
  • Inbox Summarizer — Gmail URGENT/NEEDS REPLY/FYI triage; drafts only, never sends.
  • Blog Crew / CourseFinder — topic→edited post; Tavily+YouTube learning paths, grounded only.

[02] Automations · $ ls automation/ --all

Ai-Workflow-Automations — 10 n8n (agentic Shopping Assistant w/ voice + Redis memory · Learning-Path→Docs+Calendar · Internship Applier · Job Tracker · digests) · 2 Make.com (Sheets→Gemini social pipeline · Telegram resume agent) · 1 Automation Anywhere RPA reminder bot (Try-Catch row safety). PodEase Pro ● LIVE — Lovable → n8n webhook → Gemini script → Murf TTS, idea-to-audio.


[03] Academic · $ ls team/ --scope=mine

  • CrimeSleuth (4) — trained 14-class crime-scene classifier (Colab, .pth) + YOLO Flask inference → Gemini reports. · SignatureSense (4) — integrated pre-trained Keras .h5 verifier + preprocessing/tests.
  • KMIT B.Tech CSE 2023–2027 · DSA · OS · DBMS · Networks · SE · Cloud (AWS) · Cyber Security. Extras: F5-TTS Kaggle voice-clone · FoodMunch Bootstrap page.

[04] Credentials · $ cat certifications.md

StudyComrade Data Analytics (EDA on Amazon dataset) · Outskill AI Mastermind (assistants/GPTs, Claude Artifacts, generative media, AI-built sites).


[05] Arcade · $ ./play.sh --pacman

Pac-Man eating the contribution grid

Pac-Man regenerates daily via .github/workflows/pacman.yml → output branch (abozanona/pacman-contribution-graph) — grid recolored post-generate to the maithresh.sh cyan ramp. Trigger once manually in Actions after push.

$ ./contact.sh --open

📫 maithresh.sh · LinkedIn · maithreshvaddi16@gmail.com — open to AI/ML, GenAI, Agentic AI, Automation internships. Hyderabad / remote.

exit 0 — thanks for visiting

Pinned Loading

  1. TrustRAG TrustRAG Public

    TrustRAG — local-first RAG reliability workbench: claim-level NLI verification, SHA-256 evidence audit and self-healing retrieval over hybrid dense+BM25, fully offline by default

    Python

  2. MCP_Agentic_DocuChat MCP_Agentic_DocuChat Public

    MCP_Agentic_DocuChat: a local-first reading companion for your PDFs. Ask a question and get an answer that names the exact pages behind it, so you can check the claim yourself. Works fully offline …

    Python

  3. CareerOS-Pro CareerOS-Pro Public

    CareerOS-Pro aggregates live job and internship listings from multiple job boards and company career pages, filters results with deterministic rules, ranks them using hybrid keyword and semantic s…

    Python

  4. Resume_Crew Resume_Crew Public

    Resume Crew is a Privacy-first AI resume matching for candidates and recruiters—grounded job-fit insights, skill gaps, tailored bullets, and interview prep with MLX-LM, Ollama, or Gemini.

    Python

  5. Ai-Workflow-Automations Ai-Workflow-Automations Public

    Practical AI automation workflows using LLM agents, structured outputs, tool calling, APIs, Redis memory, webhooks, and RPA across n8n, Make.com, and Automation Anywhere.

  6. CrimeSleuth CrimeSleuth Public

    CrimeSleuth is an AI-powered forensic case management system that enables investigators to create cases, upload and analyze evidence using computer vision and multimodal AI models, interact with ca…

    TypeScript