Kunal User
@kunaluser8
Entry-level AI/ML engineer building retrieval systems, neural compression, and AI-agent infrastructure.
IndiaWhat I'm looking for
I’m a second-year B.Tech student in Material Science and Engineering at IIT Kanpur, focused on applying ML to build practical AI systems. I enjoy turning research ideas into working software—whether that’s compression pipelines or agent-facing APIs.
On the AI side, I’ve contributed to open source projects like LiteLLM (AI Gateway) and MemPalace (MCP Memory Server), fixing real edge cases and shipping improvements with tests and versioned releases. I also built AgentGate, an “Auth0 for AI Agents” hackathon project that enforces Allow/Block/Step-Up policies, fetches tokens at runtime, and coordinates real-time sync with WebSockets.
I’m equally driven by performant models: in DNN Compression, I compressed a 19.7M-parameter CNN by 141.5× while keeping accuracy nearly intact, using pruning, k-means quantization, and Huffman coding. For NLP, I shipped a QA chatbot (PULPNET) with BM25+FAISS retrieval, cross-encoder reranking, and DistilBERT extractive QA, complete with a multi-turn UI and source citations.
Experience
Work history, roles, and key accomplishments
Built an HTTP proxy for AI agents enforcing Allow/Block/Step-Up policies on Gmail and GitHub calls, while eliminating credential storage by fetching OAuth tokens at runtime via Auth0 M2M. Engineered real-time 3-way sync using WebSockets (15-min expiry with auto-retry) and delivered a human-in-the-loop approval dashboard (deployed on Railway + Vercel).
DNN Compression via EEA
EEA IIT Kanpur
Dec 2025 - Jan 2026 (1 month)
Compressed a 19.7M-parameter CNN by 141.5× (47.3 MB to 0.33 MB) with only 0.37% accuracy drop using pruning, k-means quantization, and Huffman coding. Recovered 95.29% accuracy after 99.7% weight pruning via BatchNorm removal and 5-epoch cosine LR retraining, and achieved 95.65% on Fruits-360 with a dual-branch CNN fusing LBP/Canny descriptors.
MemPalace MCP Params Fix
MemPalace
Apr 2026 - Present (2 months)
Fixed an LLM agent breakage where empty/whitespace wing/room parameters incorrectly triggered validation errors despite being schema-optional. Shipped the fix in v3.3.3.
LiteLLM OVHCloud Migration
LiteLLM
Apr 2026 - Present (2 months)
Implemented OVHCloud API field migration across transformation classes, adding backward-compatible fallbacks and fixing a falsy-zero edge case (seconds=0.0). Added 6 unit tests (Greptile 5/5).
PULPNET Domain QA Chatbot
ACA IIT Kanpur
Jan 2024 - Present (2 years 5 months)
Built a domain-specific IIT Kanpur QA chatbot using hybrid BM25+FAISS retrieval, cross-encoder reranking, and DistilBERT extractive QA. Implemented token chunking (384 tokens, 64-token overlap) over 16 scraped pages and delivered multi-turn chat with source citations and confidence scoring (EM: 0.60, F1: 0.66).
Education
Degrees, certifications, and relevant coursework
Indian Institute of Technology Kanpur
B.Tech, Materials Science and Engineering
2024 -
Grade: 5.8/10
Pursuing a B.Tech in Materials Science and Engineering at IIT Kanpur (performance: 5.8/10).
RLC Public School, Dabhara
CGBSE (XII), Higher Secondary (CGBSE XII)
2023 -
Grade: 83%
Completed CGBSE (XII) at RLC Public School, Dabhara with 83%.
Kamdhenu Public School, Kirari(D)
CGBSE (X), Secondary (CGBSE X)
2021 -
Grade: 97%
Completed CGBSE (X) at Kamdhenu Public School, Kirari(D) with 97%.
Availability
Location
IndiaAuthorized to work in
Job categories
Interested in hiring Kunal?
You can contact Kunal and 90k+ other talented remote workers on Himalayas.
Message KunalFind your dream job
Sign up now and join over 250,000+ remote workers who receive personalized job alerts, curated job matches, and more for free!
