Saksham Adhikari
॥

head down grinding

Saksham Adhikari

Hi, Saksham here. I care about building AI that is reliable, measurable, and useful in the real world—from the model and mathematics underneath it to the infrastructure that makes it run.

the kites follow your cursor. pressto make them fight,for night

$ wc -l the receipts

By the numbers

The receipts, counted.

6

hackathon wins

NVIDIA, webAI, Vercel v0 × AWS, UT Law, TXST, Novo

~130k

records indexed

Cancer-care records, semantically searchable at BCRC

180k+

grants queryable

Grants-MCP, 389+ downloads on PulseMCP

~10,000

hackers beaten

Best Monetizable B2C App at H0, Vercel v0 × AWS

78% → 91%

test coverage

Agentic healthcare stack at BCRC

4.0

GPA

B.B.A. Computer Information Systems, Texas State

reading the calendar

github.com/Tar-ive · the last year

FETCHING

SunTueThuSat
updatingLessMore

$ git log --author=saksham

Professional Experience

Inference on hardware, health data at scale, and the teams in between.

May 2026 — Aug 2026
AskSLM
Jan 2026 — May 2026
AskSLM
Mar 2025 — Jan 2026
Translational Health Research Center
Aug — Oct 2025
Google
May — Aug 2025
Breast Cancer Resource Center
Dec 2024 — Aug 2025
ACM AI @ TXST
Dec 2024 — May 2025
Obvius
Aug 2024 — Feb 2025
AI4ALL Ignite
Sep 2024 — Jan 2025
Intern Nepal
Mar — Jun 2023
Dursikshya Education Network

Machine Learning Engineer Intern

AskSLM

Austin, TXMay 2026 — Aug 2026

Technologies

LLaVAquantizationDeepStreamRAGHarbor

What I did

  • Quantized custom-trained LLaVA vision models for edge devices, monitoring weight clustering for a 3x increase in image-to-text sequence recognition throughput.
  • Added Reciprocal Rank Fusion to the internal RAG pipeline, lifting retrieval precision for a 3% improvement in Mean Reciprocal Rank.
  • Engineered real-time video analytics pipelines on the DeepStream SDK, tuning frame-rate throughput and end-to-end inference latency with Harbor for systematic evaluation.

$ ls ~/projects

Projects

Rewind's dashboard: a day of tasks rescheduled by priority, with agent activity alongside

featured · TreeHacks 2026

Rewind

The first productivity system built around disruption, not planning.

6.1Madults in the U.S. are diagnosed with ADHD

But diagnosis isn't the problem. Executive dysfunction because of ADHD means that even when you know exactly what you need to do, starting feels impossible. The planning isn't the problem, it's truly when the plan breaks.

Two members of our team live with ADHD. We've tried every productivity system: Notion databases, time-blocking on Google Calendar, even gamified to-do apps. They all share the same fatal flaw: they help you plan, but they abandon you the moment something goes wrong.

A meeting runs 15 minutes over. You space out for 10. By the time you've replanned, 45 minutes are gone, and your momentum with it.

That's the exact moment we're targeting: the 15 to 30 minutes after a disruption occurs, when the plan breaks and the user is silently stuck because they have no idea what to do next.

how it works

Six agents run it like an operating system: a Context Sentinel watches calendar, email and Slack; a Disruption Detector grades what just broke; a three-tier Scheduler Kernel (long, medium and short term, on a modified multilevel feedback queue with bin-packing and task swapping) rebalances the day in seconds; an Energy Monitor matches tasks to how much focus you have left; a Profiler learns your patterns; and GhostWorker drafts the reschedule emails for you.

Fetch.ai uAgentsComposio MCPFastAPIRedisNext.jsGoogle CalendarGmailSlack

TwoBot

A two-tower GenRecSys where on-device curator agents (MLX) evaluate candidates and write personalized surfacing notes. Live A/B of a recency baseline against two-tower + MMR on a 1,300-node simulation.

MLXtwo-towerMMRagents

$ ls ~/wins

Hackathons

Six wins, and the things they turned into.

H0: Hack the Zero Stack · Vercel v0 × AWS

Giftmaxxing

Best Monetizable B2C App · ~10,000 participants

An AI gift concierge built with Kusum Bhattarai Sharma. Won $6,000+ in prizes, then shipped to the App Store off feedback from AWS engineers and judges.

iOSAWSv0recsys

$ whoami

Education

Academic background and the stack I work in.

Texas State University

B.B.A. Computer Information Systems

San Marcos, Texas4.0 GPA

Full-tuition Merit Scholar, one of 15 awarded schoolwide. AKAEF Undergraduate Launch Scholar and Merry Kone FitzPatrick Endowment Scholar.

Inference engineeringComputer visionRecommendation systemsLLM optimizationNVIDIA JetsonDeepStreamvLLMllama.cppMLXPyTorchTensorFlowPythonTypeScriptFastAPINext.jsPlaywrightMCPInference engineeringComputer visionRecommendation systemsLLM optimizationNVIDIA JetsonDeepStreamvLLMllama.cppMLXPyTorchTensorFlowPythonTypeScriptFastAPINext.jsPlaywrightMCP

coursework · keep scrolling

$ mail -s

Get in touch

Working on inference, ranking, or health data? I answer every note.

Currently seeking

Open to 2027 new grad SWE, ML opportunities.

Find me

[email protected]

Austin / San Marcos, Texas

Slide a DM

This opens your default email client with the message prefilled. Nothing is stored or sent to a server.