Loading…

Welcome.

ABHILASH ANUKU

48.7758°N · 9.1829°E

abhilashanuku
Applied AI · Stuttgart · --:--

Full-stack developer, M.Sc. Applied Computer Science at SRH Stuttgart. Building applied-AI systems — multi-agent cascades, RAG, self-hosted deployment. Open to Werkstudent and junior AI roles in Germany.

Explore my work
GitHubLinkedInEmail
01 · About

Student,
still building

Abhilash Anuku. From Guntur, Andhra Pradesh. Undergrad in Kuppam. Two years writing production Laravel and PHP at Datalense Services in India after that. Now in Germany, finishing the M.Sc. in Applied Computer Science at SRH University Stuttgart — started October 2025, expected through 2027.

I'm twenty-five. I've been writing code since 2019. I grew up between school in Guntur and a CS undergrad in Kuppam, and the work that interests me now is the same work that interested me then — building tools that make a person more capable than they were yesterday. The form has changed; the question hasn't. The projects below are how I work — one build at a time, each with a README that names its current limits.

During undergrad in Kuppam I also built and hosted the Lions Club Kuppam website and a few related civic sites, and guided junior students from my college on the code. Volunteer, unpaid; the goal was keeping local orgs online without them needing a paid agency. Letters of recommendation available on request.

Current focus: applied AI. Multi-agent systems, retrieval-augmented reasoning, MCP-native architecture, self-hosted deployment. Python for agents, TypeScript for interfaces, Postgres + ChromaDB / Qdrant for state.

Three builds run in parallel right now: LISA (a self-hosted personal AI), Dr.ABC (a medical-reasoning cascade), and Mörbius (a clinical-reasoning research project). Full numbers and stack on the cards below — each sits in the repo with a README that names its current limits.

Languages: English (fluent), Telugu (native), Hindi (fluent), German (A1, actively learning). Open to Werkstudent and junior applied-AI roles in Germany.

Coding since

0

Languages

EN · TE · HI · DE

Currently building

LISA · Dr.ABC · Mörbius

Open to

Werkstudent · Junior AI

PythonTypeScriptNext.jsReactFastAPIFastMCPHonoPostgresSQLAlchemyDrizzleChromaDBQdrantThree.js · R3FTailwind v4DockerCanva
Abhilash AnukuStuttgart · 2026
02 · Journey

The
journey

Oct 2025 — Sep 2027

now

Stuttgart, DE

M.Sc. Applied Computer Science

SRH University · Stuttgart, Germany

  • Focus areas: applied AI, distributed systems, software architecture, intelligent automation.
  • Coursework includes the GenAI module that produced Dr.ABC — medical-reasoning cascade with online residual correction.
  • Building Mörbius alongside the M.Sc. as a self-led research project — a six-agent clinical-reasoning cascade with retrieval, memory, orchestration, and adaptive inference. Not the thesis; pursued in parallel.

Earlier

Kuppam, IN

B.Tech — Computer Science Engineering

Kuppam, India

Earliest

Guntur, IN

Early learning · self-taught development

Guntur, India

03 · Skills

The
toolkit

grouped by surface — only what I've shipped code against

Languages

AI / ML

Backend

Frontend

Ops

Tools

04 · Work

What I'm on

Oct 2025 — present

now

M.Sc. student · independent builder

SRH University Stuttgart

Applied AI focus. Building LISA, Dr.ABC, and Mörbius in parallel — a personal-AI runtime, a medical-reasoning cascade, and a self-led clinical-reasoning research project. Open to Werkstudent and junior applied-AI roles in DE.

2024 — 2025

Generative-AI coursework project (solo)

SRH · Dr.ABC

Built a medical-reasoning cascade for the GenAI module: specialist agents, online residual correction, Postgres-backed error journal, vector retrieval. First benchmark pass came back at 74.5% on MedQA and 74% on MedMCQA on an internal eval — public eval pending.

May 2023 — Jul 2025

Full Stack Developer

Datalense Services · Kuppam, India

Two years of production code before Germany. CRM and operations apps in Laravel, PHP, MySQL, and Bootstrap 5 — paying clients, real release cycles, end-to-end delivery. One of the three entities Abhilash worked with under a shared owner.

2023 — 2025

Full Stack Developer

Digital Kuppam · Kuppam, India

Sister organisation to Datalense — same owner, separate company, parallel work.

2023 — 2025

Freelance Developer

Kada Kuppam Development · Kuppam, India

Smaller local builds outside the agency workload. NGO sites, lightweight CRMs, and automation tooling for organisations that could not pay agency rates. Same owner network as Datalense + Digital Kuppam.

05 · Projects

Things I'm
building

hover to flip

LISA

Active · v0.2

Self-hosted personal AI with persistent memory, voice, and an MCP-native tool layer.

PythonFastMCPSQLAlchemyNext.jsBetter Auth

LISA

open ↗
  • — 4 FastMCP capability servers — chat / RAG / voice / features — on dedicated ports
  • — 5-provider chain: Groq → NVIDIA NIM → Cerebras → Google AI Studio → Ollama (local fallback)
  • — Shared SQLite + ChromaDB, single-host Docker deploy behind Caddy
  • — Next.js admin panel + Tauri desktop overlay for the daily driver

Dr.ABC

Shipped · coursework project

Medical-reasoning agent cascade with online residual correction.

PythonFastAPIDrizzleQdrantPostgres

Dr.ABC

open ↗
  • — 74.5% on MedQA · 74% on MedMCQA — first benchmark pass on an internal eval set, single-run. Public eval pending.
  • — Cascade of specialist agents — each owns a narrow lane (retrieval, ranking, reasoning, verification, citation, synthesis)
  • — Postgres-backed error journal via Drizzle — every wrong answer is logged with the trace that produced it
  • — Qdrant for vector retrieval, with the residual-correction loop running over the journal

Mörbius

Self-led research · in flight

Self-led research project — six-agent clinical-reasoning cascade over a PubMed case-report library.

BunTypeScriptReact 19HonoDrizzlePostgres

Mörbius

open ↗
  • — 6-agent cascade over 532 PubMed case reports — each agent owns one reasoning step
  • — Knowledge graph as the context layer between retrieval and the reasoning loop
  • — Multi-provider runtime: NVIDIA NIM, Anthropic, HuggingFace, Ollama — local Ollama fallback if cloud providers fail
  • — 588/588 tests green at last run — the cascade has 588 tests that must pass before any merge
07 · Writing

From
LinkedIn —

the same posts, here in long-form.

2026-05-12

drafting

Dr.ABC at 74.5% MedQA: what the number does and doesn't say

First benchmark pass on Dr.ABC came back at 74.5% on MedQA and 74% on MedMCQA — internal eval set, single run, public eval pending. Two questions worth answering before anyone celebrates: what kind of questions the cascade gets right, and where it falls apart. The cascade is six agents — retrieval, ranking, reasoning, verification, citation, synthesis — with a Postgres-backed error journal driving the residual-correction loop. Strength sits in single-step recall; multi-hop is where it slips.

2026-04-28

drafting

Three projects in parallel: how I run LISA, Dr.ABC, Mörbius

LISA is the personal AI runtime — four FastMCP capability servers on ports 7878–7881, a five-provider chain (Groq, NVIDIA NIM, Cerebras, Google AI Studio, Ollama). Dr.ABC borrows LISA's provider chain so benchmarks fall back to Ollama when cloud quota runs out. Mörbius runs on its own Bun + Hono stack with NIM and Anthropic, ingesting PubMed case reports into a knowledge graph. The shared runtime is the only reason the calendar works.

drafting

drafting

Why LISA's four MCP servers don't merge into one

Capability boundaries, port discipline, and the failure mode I avoided by keeping chat, RAG, voice, and features on separate processes. One MCP, one responsibility — ports 7878 through 7881, each crash-isolated. Restarting the voice server doesn't take down chat.

More notes on linkedin.com/in/abhilashanuku — source on github.com/AbhilashAnuku.

08 · Contact

Reach
out

Open to Werkstudent and junior applied-AI roles in Germany. I reply within 48 hours.

Emailabhilashanuku14@gmail.comWhatsAppchat on WhatsAppLinkedInin/abhilash-anukuGitHub@AbhilashAnuku
LocationStuttgart, Germany