About

Hi, I'm Jigar Kanakhara, an MS Computer Science student at Boston University building multi-agent LLM systems and doing AI safety research. I ship to production and win hackathons doing it: five wins in one year across agentic AI, medical imaging, and civic tech.

I am a Research Assistant with Prof. Aaron Mueller studying sycophancy in instruction-tuned LLMs, and recently a Teaching Assistant for CS460 Database Systems. Open to full-time roles and Spring 2027 internships in AI, ML, and full-stack engineering.

Focus areas

  1. 01Agentic AI
  2. 02LLM fine-tuning
  3. 03Influence functions
  4. 04vLLM inference
  5. 05AI safety
  6. 06Vision Transformers
  7. 07Full-stack

Interested in

  1. 01AI safety and alignment
  2. 02Multi-agent systems
  3. 03Mechanistic interpretability
  4. 04LLM efficiency
  5. 05Applied ML

Research

Sycophancy in LLMs · Influence functions · EV route planning, in press with Springer

2024 — Present

Experience

Research, teaching, and applied AI across Boston University and industry.

  1. May 2026 — Present

    Research Assistant at Boston University, Prof. Aaron Mueller

    Boston, MA

    AI safety research on sycophancy in instruction-tuned LLMs.

    Trace sycophantic behavior back to specific post-training examples with influence functions (EK-FAC).

    Fine-tune Olmo 2 7B Instruct on Tulu 3 with adversarial confounders: surface-similar agreement, polite decoys, warm corrections.

    Validate causally through leave-out retraining across seeds while preserving MMLU and GSM8K.

    Influence functions / LLM fine-tuning / AI safety

  2. May 2026 — June 2026

    Teaching Assistant, CS460 Database Systems at Boston University, Prof. George Kollios

    Boston, MA

    Graduate database course covering SQL, indexing, query evaluation, and transactions.

    Designed programming assignments, the final project, and reference solutions.

    Wrote grading rubrics and evaluated submissions against them.

    SQL / Transactions / Query evaluation

  3. Jan 2025 — May 2025

    Data Analyst at Strategic ERP

    Ahmedabad, India

    Built an internal AI assistant so teams could query the ERP in natural language.

    Fine-tuned Llama 3 8B Instruct with LoRA on CRM and HRMS docs, HR policies, and query patterns for 4 enterprise clients.

    Deployed as a REST endpoint for sales, HR, and support teams.

    Fixed recurring data inconsistencies at source and cut error tickets by 25%.

    Llama 3 / LoRA / REST

  4. May 2024 — Aug 2024

    Research Assistant at IIIT Vadodara, Dr. Naveen Kumar

    Gandhinagar, India

    EV route planning with a Fuzzy Genetic Algorithm for charging-station selection.

    Selected charging stops across multi-stop routes, benchmarked against greedy and shortest-path baselines for 26% lower cost.

    Built a Python simulator with pandas-based trip analysis.

    Presented at CTSEM 2025 and in press with Springer Lecture Notes in Civil Engineering.

    Genetic algorithms / Python / Optimization