● portfolio.md ×

○ editorial.html ×

whoami

Aditya Dasika_

role: AI/LLM Engineer at: Phenom People loc: Hyderabad, IN exp: 4 yrs

// tagline Building production AI systems — from "LLM fine-tuning" to "physics-informed neural networks".

I architect agentic conversational pipelines, fine-tune LLMs at scale on AWS, and build Graph RAG systems for enterprise. On the side I implement PINNs from scratch in PyTorch — heat, Burgers', Schrödinger. Mechanical engineering by training, self-taught into production ML.

akhil.dasika47@gmail.com ⏎ github.com/adityadasika21 linkedin/in/adityadasika21 download resume.pdf ⌘D

97%

// nlu accuracy

87% → 97% via mt-bert

5.72 rps

// prod throughput

qwen3.5-2b on g5.xlarge

98.6%

// success rate

at peak load

10k+

// monthly txns

blueleaves pos · live

$ cat about.md

25 lines · markdown

about.mdmarkdown

1# aditya dasika
2
3> AI/LLM engineer at Phenom People. I architect
4> 12-node LangGraph agentic pipelines, fine-tune
5> LLMs at scale on AWS, and build Graph RAG systems.
6
7> On the side I implement physics-informed neural
8> networks from scratch in PyTorch — heat, Burgers',
9> Schrödinger.
10
11> Mechanical engineering by training, self-taught
12> into production ML — I think about AI systems
13> differently because I came in through physics,
14> not through web apps.
15
16## vitals
17- based: "Hyderabad, IN"
18- role:  "AI / LLM Engineer @ Phenom People"
19- since: "April 2025"
20- prior: "Full-stack @ Blueleaves Farms · 2.5 yrs"
21- edu:   "B.E. Mechanical → self-taught ML"
22- stack: "LangGraph · PyTorch · vLLM · AWS"
23- status: "open to AI/ML engineering roles"
24- resume: "Aditya_Dasika_Resume.pdf" ↓

$ ls -la work/

2 entries

phenom.py apr 2025 → present · hyderabad · enterprise hr-tech

Phenom People / AI · LLM Engineer

chatbotCXAgent · 12-node LangGraph pipeline

Architecting an enterprise conversational AI: 12-node LangGraph pipeline with discrete owners — Concierge, Intent Classifier, Dialogue Manager, NLG, CTC, and seven more. QLoRA fine-tuned Qwen3.5-2B on 4,569 production records, deployed to vLLM on AWS EC2 g5.xlarge.

Built the eval framework end-to-end (pass@k, GPT-4.1-mini judge rubrics across all 12 nodes) and lifted intent accuracy from 87% to 97% with a multi-task fine-tuned ModernBERT (intent + NER, joint heads). Graph RAG pipelines on FalkorDB for structured retrieval.

87 → 97%

// nlu accuracy

multi-task ModernBERT (intent + BIO NER)

5.72 rps

// load tested

qlora qwen3.5-2b · vllm · g5.xlarge

98.6%

// success rate

at peak rps · mixed traffic

7 → 78%

// test coverage

unit + node + llm-judge

// production debugging — 3 i'm proud of
parallel tool calls in agent loop generated spurious params
isolated invocation context per branch · pipeline stable
JSON schema $ref recursive loops crashed certain payloads
flattened schemas at boundary · zero crash since
vLLM + Qwen mamba+attention hybrid: undocumented at the time
merged weights + --enforce-eager · shipped to prod

LangGraph LangChain FastAPI Redis PostgreSQL FAISS FalkorDB ModernBERT Qwen3.5-2B vLLM PyTorch Langfuse Kafka Unsloth QLoRA / NF4 AWS EC2 g5.xlarge SageMaker

blueleaves.tsx sep 2022 → feb 2025 · sole developer

Blueleaves Farms / Full-stack Engineer

POS system · 5 locations · 2 brands · web + mobile

Sole developer, end-to-end. Multi-location POS from scratch — billing & checkout, SKU-grouped inventory (variants collapsed into single SKUs), cash handling, Razorpay payments, thermal receipt printing, role-based accounts for 20 staff, and a React Native mobile POS variant.

Owned the schema, the React + MUI web app, the Expo mobile app, the Python Cloud Functions backend on GCP, and the on-site network and hardware setup at all 5 locations. 3-hour solo SLA — no escalation path. 2 years live.

10k+

// monthly txns

2 yrs in prod

200+

// daily peak orders

concurrent across locations

// city locations

multi-location stock sync

// staff accounts

rbac · cashier / admin

// scope i owned, top to bottom

product: billing/checkout · SKU-grouped inventory · cash handling · Razorpay · thermal receipts · rbac (20 staff)

reach: react + MUI web · react native + expo mobile · multi-location stock (5 cities) · 2 brand catalogs (30 SKUs)

platform: schema · firestore + auth + hosting · python cloud functions on GCP · on-site network & hardware setup

solo on-call: 3-hour SLA · no escalation path · 2 years live, still running

React Material UI React Native Expo Firebase Firestore Firebase Auth Cloud Functions (Python) GCP Razorpay SDK thermal printer SDK Photoshop Blender on-site sysadmin

$ ls projects/

4 entries · hover to reveal core idea

pinns.ipynbP/01

Physics-Informed Neural Networks

PyTorch · from scratch · 3 PDEs

Implemented automatic differentiation for physics residuals — no PDE-specific frameworks. Heat, Burgers', Schrödinger. Two-phase optimization (Adam → L-BFGS), complex-valued nets where the physics demands it.

~3%

heat eq. err

Raissi parity

burgers'

8.3e-5

schrödinger

PyTorch autograd L-BFGS complex nets

∂u/∂t = α · ∂²u/∂x²

karpathy-gpt.pyP/02

Karpathy GPT — from scratch

transformer · autograd · 5k steps

Full transformer, no shortcuts. Q/K/V projections, q @ k.T, the actual VJP graph through create_graph, LayerNorm, teacher forcing. The point wasn't to ship — it was to internalize.

4.39 → 1.86

loss

5,000

steps

tinyshake

corpus

PyTorch multi-head attention VJPs teacher forcing

attn(Q,K,V) = softmax(QKᵀ/√d)·V

mmbert.pyP/03

mmBERT — multi-task NLP

ModernBERT · 2 task heads · uncertainty weighting

ModernBERT backbone, two heads: 4-class intent + BIO NER. Uncertainty-weighted loss (Kendall et al.). Diagnosed validation loss divergence — fixed with differential learning rates, stratified splits, label smoothing.

intent + NER

heads

σ-weighted

loss

prod

candidate

ModernBERT multi-task HuggingFace PEFT label smoothing

L = Σᵢ (1/2σᵢ²)·Lᵢ + log σᵢ

indexnotes.cypherP/04

IndexNotes — paper knowledge graph

Neo4j · Ollama · PDF → graph

Reading a paper isn't enough — I wanted to see how ideas connect. Neo4j-backed concept graph: ~40 concepts, 17+ relationships across three papers (Uni-HamGNN, PINN Part I & II). PDF in, structured graph out.

~40

concepts

17+

relations

papers

Neo4j Ollama PDF parsing graph schema

G = (V_concepts, E_relations)

$ cat skills.json

4 clusters · hover any value

skills.jsonjson

{
  "llm_genai": [
    "LangGraph", "LangChain", "RAG", "Graph RAG",
    "QLoRA fine-tuning", "vLLM", "LLM deployment",
    "prompt engineering", "Langfuse", "multi-agent systems"
  ],
  "ml_dl": [
    "PyTorch", "Transformers", "BERT/ModernBERT",
    "PINNs", "multi-task learning", "uncertainty weighting",
    "FAISS", "vector search"
  ],
  "infra_backend": [
    "FastAPI", "Redis", "PostgreSQL", "FalkorDB",
    "Neo4j", "Kafka", "AWS EC2", "SageMaker", "Docker"
  ],
  "tools_frontend": [
    "React", "Unsloth", "HuggingFace PEFT", "Ollama", "Git"
  ]
}

$ mail aditya

three ways

Let's talk.

// open to AI/ML engineering roles · Hyderabad / remote

// email · best

akhil.dasika47@gmail.com

Aditya_Dasika_Resume.pdf ↓