RETRIEVAL
RAG on Sentence-BERT + ChromaDB — grounded answers only. Hallucination-free is the floor, not the flex.
AI/ML ENGINEER — VADODARA, IN — OPEN TO WORK
MANAV AGHERA
I'm Manav, and I've spent 2+ years answering one question: how do you make AI reliable enough to trust? Answer: deterministic RAG, evals before everything, and systems that ship. Now I'm chasing the next question on CHRONUS — AGI, grief tech, memory that outlives us. Anthropic-certified. Top 1% on Kaggle, twice.
SEE MY WORK ↗THIS PAGE RUNS ON YOUR SCROLL — FOUR PINS, ONE HANDLER
OPPORTUNITY
IS EVERYWHERE.
EXECUTION
ISN'T.
TRYING TO BUILD WHAT DOESN'T EXIST YET — AI • GENAI • LLMS • ML • AGI • GRIEF TECH
HOW I BUILD — A SESSION, START TO SHIP
LOCK IN.
SHIP OUT.
00:00SPARK HITSIDEA LOCKED, TABS OPEN, MUSIC ON
08:00FLOW STATEWORLD OFF, KEYS CLACKING, COMMITS FLYING
19:00BOSS FIGHTONE BUG — 47 STACKOVERFLOW TABS DEEP
29:30CLUTCH FIXTESTS GREEN, HEART RATE NORMAL AGAIN
40:00VICTORY LAPDEPLOYED, VIBES INTACT, GRASS TOUCHED
BUILD 042SPEC
UPTIME SINCE FIRST DEPLOY
00D 00H 00M 00S
EVAL CASES 214/214 ALL GREEN
TOKENS BURNED: 5,214,397,102 EST.
EVERY BUILD RUNS THE RED LINE.
EVERY STEP MEASURED, EVERY LINE TRACED — SCROLL AND SEE
01 — ABOUT ME
THE PERSON BEHIND THE BUILDS.
I MAKE MACHINES READ, REASON AND SEE.
I'm an AI/ML engineer who doesn't stop at the model. 2+ years across machine learning, deep learning and GenAI — I design RAG pipelines that never hallucinate, build neural networks that actually learn, and wire multi-agent systems that fix themselves when they fail. Fine-tuning, inference optimization, vision on the edge — the full stack of making AI real.
Currently finishing my B.Tech in CS at Parul University. Anthropic Academy-certified in MCP + RAG etc , twice Top 1% on Kaggle — and a year spent break-testing pre-release builds for Riot Games taught me how real software breaks before millions touch it. Between builds I ship open-source PRs through GDSC / GITHUB and chase the questions with no answers yet: AGI, grief tech, memory that outlives us.
- LOCATIONVADODARA, IN
- EDUCATIONB.TECH CSE — PARUL
- FOCUSNEURAL NETS · LLM SYSTEMS · RAG · MULTI-AGENT · VISION · AGI · GRIEF TECH
- STATUSOPEN TO WORK
02 — WHAT I DO
THREE DEEP FIELDS.
A
B
AGENTS
LangGraph pipelines that catch their own failures and fall back without drama — scored by evals, not vibes.
C
VISION
YOLOv8 squeezed into TFLite until it runs offline on the cheapest Android in the field.
03 — PROJECTS
THINGS I'VE BUILT.
CHRONUS
Memory-grounded AI research platform — currently in RESEARCH PHASE: deterministic RAG (Sentence-BERT + ChromaDB) is live today, while I design the next chapter — AGI, grief tech (AI that helps preserve the people we love) and advanced memory architectures. Roadmap: homomorphic-encrypted vector search, quantum NLP, neuromorphic sandboxes.
AUTONOMOUS-CONTENT
4-agent LangGraph pipeline — research → writer → validator → publisher — with automatic failure detection and model fallback. Scored 78.7/100 on automated content validation; six technical articles published.
DERMALENSE.AI
Browser-first skin-health computer vision experiment — built in JavaScript, fast and lightweight, runs entirely in the tab. No servers, no uploads, straight inference.
MCP-DEMO
Hands-on Model Context Protocol playground — servers, tools and resources built from the Anthropic Academy curriculum. The reference implementation behind my MCP certification.
FASAL AI
Real-time crop-disease detection (YOLOv8 · TensorFlow) with Hindi and Gujarati voice input via Whisper — offline-first TFLite for low-RAM Androids.
TFLITE
LLM-NEWSHUB
Scrapes news and social sources, clusters stories with TF-IDF + K-Means, generates LLM articles with audio, image and video summaries — React + FastAPI, Dockerized deployment.
04 — EXPERIENCE & EDUCATION
THE RECORD SO FAR.
2024 — 2025
RIOT GAMES — BETA TESTER
Remote. A year spent breaking 50+ pre-release builds before public launch — learning how real software fails at scale, one reproducible bug report at a time.
50+ BUILDS
JUL — AUG 2023
BRAINY BEAM — DS/ML INTERN
Ahmedabad. End-to-end ML pipeline on 40,000+ banking records — best model hit 92% accuracy (F1, AUC-ROC evaluated); preprocessing cut ~30% with automated cleaning scripts.
92% ACC
EDUCATION
2025 — 2027
PARUL UNIVERSITY — B.TECH CSE
Vadodara. Computer Science Engineering — deep-diving ML systems, LLM architecture and everything shipping on this site. Expected May 2027.
EXP. MAY 2027
APR 2025
NG PATEL POLYTECHNIC
NG PATEL POLYTECHNIC
— DIPLOMA CE
Bardoli. Diploma in Computer Engineering — graduated with a CGPA of 8.4 and a habit of finishing what I start.
CGPA 8.4
05 — TOOLS OF THE TRADE
THE STACK.
EVERYTHING BELOW HAS SHIPPED SOMETHING — NO TUTORIAL STACKS HERE.
ML / DEEP LEARNING
Python, TensorFlow, TensorFlow Lite, PyTorch, Scikit-learn, YOLOv8, Sentence-BERT, Whisper, ONNX Runtime, Hugging Face Transformers
MODELS TRAINEDNLP / GENAI / LLMS
LLM Architecture & Inference, Model Fine-Tuning, RAG Pipeline Design, Prompt Engineering, Agentic Systems, LangChain, LangGraph, ChromaDB, Vector Databases, Llama 3, Ollama
SYSTEMS SHIPPEDPROGRAMMING & WEB
Python, TypeScript, JavaScript, SQL, React.js, Flask, FastAPI, REST APIs
INTERFACES BUILTDATA, TOOLS & MLOPS
Pandas, NumPy, Docker, Git, GitHub, Jupyter Notebook, BeautifulSoup4, MLflow
DEPLOYS TRACKED
06 — SIDE QUESTS
SHIPPED BETWEEN THE MAINLINES.
COMPETITIONS, CONTRACTS, COMMITS AND COURSES THAT DIDN'T FIT THE MAIN QUEST LOG.
Q1
AUTONOMOUS AGENT KAGGLE
Top 1% of 570 teams — agent-behavior prediction on Kaggle's Autonomous Agent Prediction (Beta). End-to-end: EDA to ensemble.
TOP 1%
Q2
PREDICTING STUDENT HEALTH RISK
Top 1% of 3,355 teams — Playground Series S6E7. Full tabular pipeline: feature engineering, CV strategy, submission hold.
TOP 1%
Q3
RIOT GAMES BETA TESTING
A year of breaking things professionally — 50+ pre-release builds play-tested before public launch. Found the bugs so players didn't.
50+ BUILDS
Q4
GDSC OPEN SOURCE
Pull requests and issue fixes shipped to open-source repos through Google Developer Student Club. Small diffs, real impact.
OSS
Q5
SELF-TAUGHT, RECEIPTS ATTACHED
AI/ML fundamentals built the open way — Microsoft Learn and Google courses, open-source docs, and more YouTube lectures than any university syllabus could hold. No classroom, just reps.
SELF-TAUGHT
07 — PAPER TRAIL
RECEIPTS, CERTIFIED.
P1
MCP & RAG CERTIFICATION — ANTHROPIC ACADEMY
Model Context Protocol + Retrieval-Augmented Generation, officially certified.
ANTHROPIC
P2
TOP 1% — KAGGLE AUTONOMOUS AGENT PREDICTION (BETA)
Agent-behavior prediction against a 570-team field.
KAGGLE
P3
TOP 1% — KAGGLE PREDICTING STUDENT HEALTH RISK
Playground Series S6E7 — held rank across a 3,355-team field.
KAGGLE
P4
SOFTWARE ENGINEERING VIRTUAL EXPERIENCE — JPMORGAN CHASE & CO.
Forage job simulation — engineered to enterprise spec.
JPMORGAN
P5
GDSC — OPEN SOURCE CONTRIBUTOR
The official record: PRs merged, issues resolved, repos credited.
GDSC
08 — QUICK ANSWERS
THE FAQ.
FAQ 01 — Are you open to remote work or relocating?
Both. Fully remote, hybrid or on-site — based in Vadodara, India (IST), comfortable overlapping US/EU hours, and open to relocating for the right role.
FAQ 02 — What roles are you looking for?
AI/ML Engineer, ML Engineer, GenAI/LLM Engineer — internships, full-time or research collaborations. If it involves shipping intelligent systems, we should talk.
FAQ 03 — When can you start?
B.Tech wraps May 2027 — but internships, part-time roles and remote collabs fit alongside studies. I'm used to shipping while enrolled.
FAQ 04 — Do you take freelance or contract projects?
Selectively. LLM systems, RAG pipelines, multi-agent builds and edge CV deployment — if it matches the work on this site, send the brief.
FAQ 05 — What's the fastest way to reach you?
Email — manavaghera@yahoo.com. I reply fast. LinkedIn works too, but email skips the queue.
09 — CONTACT
GOT A SYSTEM TO BUILD?
🟢 STATUS: OPEN TO WORK — REMOTE · RELOCATION · IMMEDIATE
GOT A SYSTEM TO BUILD?
LET'S TALK.
Actively open to AI/ML roles — internships, full-time, remote or relocation. Recruiters: the FAQ above has the logistics; the resume button has the receipts. Email me and hear back within 24 hours.
TOKENS BURNED: 5,214,397,102EST.
·
LAST COMMIT: RECENTLY
·
COMMITS THIS YEAR: 847
·
PUBLIC REPOS: 24
·
GITHUB ↗