01
An Adaptive Hybrid Quantum–Classical Framework for Test Suite Minimization via QAOA
first authoracceptedSoftware testing · quantum optimization
ICSTA-tier
Field notes — Delhi, India · since 2023
Lakksh
Sharma
I take apart systems that don't work as well as they should.
I find where the bottleneck actually is — then build the thing that fixes it. Some of that is research; some of it ships. Here's what I've built, a small game about me, or just say hello.
✳01 Who
The shape of the
problem I keep chasing
Six domains in two years — healthcare, maritime routing, judicial scheduling, brain signals, agents. On paper that looks scattered. It isn't. Underneath, it's the same problem wearing different clothes: optimization and structured data under uncertainty.
I'm in my third year of Computer Engineering at Thapar Institute. I split my time between research and shipping — quantum-classical hybrids and multi-agent methods on one side, real products with numbers attached on the other. Each one teaches me how to do the other better.
✳02 Interactive
Plot the constellation
Eight stars are drifting below — each one is a fact about me. Move the marker over a star to plot it, and a line draws to the last one you caught. Trace all eight and you'll have the whole picture.
PLOTTED 0 / 8
0%
Care to map me?
No timers, nothing to lose — just guide the marker into the stars and read what comes up.
move with your mouse · ↑ ↓ ← → works too
There's the whole sky.
The long version is below — the research, what I've shipped, and where I've been.
✳03 Research
Papers, in flight
First-author work across software testing, NLP and multi-agent learning — accepted and under review at A-tier and top ML/NLP venues, written before finishing undergrad.
02
Neuro-Symbolic Decomposition to Mitigate Content Bias in Syllogistic Reasoning
first authoracceptedTop-20 globally on the shared task
SemEval-2026@ ACL 2026
03
Decodable Is Not Steerable: Counterfactual Controls for Mechanistic Probing of LLM Reasoning
first authorunder reviewInterpretability · LLM reasoning
EMNLP 2026
04
CPSS: Channel-Predictive Semantic Scheduling via Factorized Multi-Agent Reinforcement Learning
first authorunder reviewMulti-agent RL · scheduling
NeurIPS 2026
05
Regime-Decomposed Federated Quantum Experts for Cloud Workload Prediction
first authorunder reviewFederated learning · quantum ML
ECML-PKDD2026
✳04 Built
Things that shipped
Not demos — each one carries real numbers: latency, accuracy, throughput. Public work from github.com/lkksharma.
01 — healthcareWinner · MANIT × Thapar Innovate 2025
SAHAJ
a platform that makes patients, doctors and pharmacies share one workflow
OCR prescription digitization, generic-alternative price analytics, 3D hospital routing and WebRTC telehealth — knitted together so three very different users meet in the same place.
92%OCR accuracy
500+pharmacy SKUs
1,000+async tasks / hr
80%patient savings
PythonTensorFlowDjangoReactFirebaseMongoDBWebRTCCelery · Redis
02 — maritime
Raahguzar
picking a ship's route out of ten thousand
A hybrid PSO–GA optimizer that searches 10,000+ candidate routes per voyage, plus an ML crew-assignment model trained on 2,500+ real voyages.
45→8 minscheduling
2,500+voyages
PythonPSO · GADjangoNext.js
03 — courts
Nya-Alaya
an agent that schedules a courtroom
RAG over 1,200+ legal documents with a priority-aware optimizer that places 100+ cases in under two seconds — deployed on AWS for 50+ concurrent users.
14d → 1.5durgent backlog
200+cases / day
PythonDjangoReactRAGAWS
✳05 Path
Where I've been
Dec 2025 — Jan 2026 · Gurugram
SDE-I Intern, Agentic AI Systems
Ignited Wings Technology (AI LifeBOT)
- Shipped 8+ core modules of a voice-first agentic-AI marketplace serving 20+ agents — discovery, orchestration, deployment.
- Brought task-routing latency from 1.8s down to under 600ms, and added 12 voice-led execution paths.
Jul — Aug 2025 · Gurugram
AI Intern, Computer Vision
Rebec Technologies
- Led a 5-person team on a real-time badminton drill-analytics pipeline running at 30 FPS.
- Pose-based classification of 6 drill types over 40+ hours of footage — coach review dropped from 50 to 20 minutes a session.
Ongoing · Patiala
Undergraduate Research Fellow
Thapar Institute of Engineering & Technology
- Working where quantum machine learning meets optimization — the source of most of the papers above.
✳06 Toolkit
What's on the bench
Languages
PythonC / C++SQLJavaScriptHTML / CSSML & AI
PyTorch · TensorFlow · Kerasscikit-learn · NumPy · PandasQuantum ML · QAOAMulti-agent RLFederated learningOpenCV · MediaPipeNLP · RAGWeb & systems
Django · Flask · RESTNode · ExpressReact · Next.jsFirebase · MongoDBWebRTCCelery · RedisPlatforms
AWSDockerGit · GitHubPostmanAndroid StudioArduino✳07 The record
A few numbers I'm fond of
- 0 / 10
- CGPA at Thapar Institute (9.9 AGPA in the scholarship year)
- 0ms
- task-routing latency, down from 1.8 seconds
- 0%
- OCR accuracy on real prescriptions in SAHAJ
- 0fps
- real-time pose analysis on live video
-
Merit Scholarship — ₹5,20,000Thapar Institute, for academic performance
-
Winner — MANIT × Thapar Innovate 2025for building SAHAJ
-
2nd place — Thapar HackTech 202524-hour AI hackathon
-
Top-20 globally — SemEval-2026shared task at the ACL workshop
-
4th — MLH HACKTU 6.0plus GSSoC '24 open-source contributor
-
92.6% — Higher Secondary (CBSE)Bharti Public School, Delhi
✳08 Say hello
I read every email. If you've got a hard problem, a lab, or just a good question — I'd like to hear it.
Built by hand in Delhi — one HTML file, no frameworks, no templates, usually after midnight. Set in Newsreader & Hanken Grotesk. Drag your cursor across the sky; the stars notice.
— Lakksh